Quantitative Easing (QE) and biases in research

Do asset purchases by central banks raise economic output and inflation? An interesting new paper finds an affirmative answer, but also – and this is the main point – that the size of the effect depends on whom you ask. If you ask central bankers, they will tell you that the effect of QE is large. If you ask independent academic researchers, they will tell you it is considerably smaller. This difference indicates that central-bank research on this topic might be biased. It also indicates that Quantitative Easing is probably not as effective as we are told.

One of the defining characteristics of financial markets since the financial crisis in 2008 is the use and influence of “unconventional policy tools” by central banks. As monetary policy rates have been close to zero, central banks have been unable to stimulate the economy via even lower interest rates. Instead, central banks have started purchasing financial assets, mainly government bonds. These alternative policy tools are labelled “asset purchases by central banks”, or simply “Quantitative Easing (QE)”.

Quantitative Easing increases the demand for government bonds, thereby raising their price and bringing down their yields. When yields on government bonds fall, other yields in the economy, such as yields on mortgage bonds, fall, too. This should promote economic activity and raise inflation, central banks argue.

Quantitative Easing is not uncontroversial, though. In several of my posts (link, link, and link), I have argued that it raises other asset prices in the economy, such as stock prices. Some fear that this induces bubble-like behavior in asset prices. Also, QE might distort signals from asset prices, causing unclear signals from prices about the underlying state of the economy and financial markets. In addition, by raising other asset prices, Quantitative Easing might contribute to increasing inequality, as financial assets are typically held by the already wealthy. On the other hand, if QE helps promoting economic activity, it helps reducing unemployment among low-income groups, which should reduce inequality, central banks argue in return (link). In the end, then, to justify Quantitative Easing, it should have a sizeable impact on inflation and output, outweighing the potentially negative effects on other parts of the economy.

Many papers have analyzed the effects of quantitative easing. A brand-new paper (link) summarizes these analyses and asks the question whether results are more positive when central bank economists analyze QE. Given that central banks influence public opinion, the latter question is important when we evaluate the most significant policy intervention during the last decade.

There are two reasons why I think this paper is particularly interesting. First, it summarizes research on QE in a neat way. It concludes that QE is effective, but not as effective as we are often told. Second, it emphasizes the importance of independent academic research. As the faculty representative on the Board of Directors at Copenhagen Business School, stressing the importance of academic research, I find this to be an important conclusion, too.

To avoid any misunderstandings about my own view, let me stress two things before getting to the results.

First, I believe that targeted central bank intervention can be useful. In my last post, I describe one monetary policy intervention that clearly fulfilled its goal (link). During a crisis, if markets are malfunctioning, there can be good reasons for policy interventions. On the other hand, I am skeptical towards the view that the advantages of endless asset purchases by central banks outside crisis periods outweigh their disadvantages. This paper indicates that QE is less powerful than central bankers tell us, lending some support to this view.

Second, my point here is not to say that central bank research is suspicious in general. On the contrary, I strongly recommend central banks to invest in economic research. I believe that better decisions are taken when based on solid academic analyses. So, central banks should be encouraged to invest in research, but their own evaluations of their own actions are probably not unbiased.

The study

Brian Fabo, Martina Jancokova, Elisabeth Kempf, and Lubos Pastor (link) study 54 analyses, written/published between 2010 and 2018, of the effects of quantitative easing in the US, the UK, and the Eurozone. 57% of the papers have been published in peer-reviewed journals. 60% of the authors are affiliated with central banks.

Lubos Pastor and co-authors collect estimates of the effect of QE on economic output and inflation across the 54 studies and report the average (and median) effects. They also investigate whether the reported effects are different if a study is conducted by central bank researchers.

Bias in central bank research

Pastor et al. list five reasons why central bank research might be biased (directly taken from the paper, page 2, here):

  • ”First, the economist may worry that the nature of her findings could affect her employment status or rank. Is she less likely to get promoted if her findings dent the bank’s reputation? Could she get fired?”
  • ”Second, the economist may be unsure whether her research will see the light of day. Bank management could in principle block the release of studies that find the bank’s own policy to be ineffective, or to have undesirable side effects.”
  • ”Third, the economist may suffer from a confirmation bias (Nickerson (1998)). A central bank employee may believe a priori that the bank’s policies are effective, and she may select evidence supporting her prior.”
  • ”Fourth, the economist may care about the bank’s reputation.”
  • ”Finally, the economist may care about her own reputation if she is senior enough to have participated in the formation of the bank’s policy.”

The findings

The main findings of the paper are collected in this graph:

Source: Fabo, Jancokova, Kempf, and Pastor (2020)

Pastor et al. report that a QE-program at its peak, i.e. when a QE program has its maximum effect, on average (across the 54 studies) raises GDP by 1.57% and the price level in the economy by 1.42%, as indicated by the blue columns in the graph (”Average all studies”). This seems to be relatively large effects, I would say.

Do these effects depend on the affiliations of researchers? Pastor and co-authors find that central-bank affiliated researchers report significantly larger effects. If counterfactually changing the share of central bank researchers in a study from 0% to 100%, the peak effect on output is estimated to be 0.723%-points larger and the peak effect on the price level 1.279%-points larger. I illustrate this in the figure above as the “Effect of CB authors”. Compared to the overall average effect, the effect of central-bank authorship is large. For the price level, going from 0% to 100% central-bank authorship almost corresponds to the total average estimated effect of QE on price levels across all 54 papers . In other words, if you think central bank estimates might be biased, the ”true” average effect of QE programs is considerably smaller than the estimated overall average effect.

Pastor and coauthors report the average point estimate of all papers and the additional effect of central bank affiliation, as indicated in the figure above. They do not report the average estimate from papers written by central bank authors, respectively written by academic authors. I asked Lubos Pastor about this. In private email correspondence with Lubos and Elisabeth Kempf, they inform me that papers written by central bank authors (defined as papers with at least one central bank author) estimate a 1.752% peak effect on output. Papers written by academics (defined as papers with zero central bank authors) find a considerably smaller effect, 0.996%. For inflation, the peak effect on output is 1.791% for papers written by central bank affiliated authors. Academics estimate a much smaller effect, only 0.545%. In spite of massive asset purchases (we are literally talking trillions of dollars, euros, and yen), the average effect on inflation is small, at 0.5%, academics report. Less than a third of what central bank affiliated researchers report.

As academics, we are not only interested in the size of the coefficients/effects, but also whether effects are statistically different from zero. Pastor and coauthors report a striking finding here. While all papers written by central bank researchers find that QE has a statistically significant effect on output/inflation, only 50% of papers written by independent academics find significant effects.

Finally, Pastor and co-authors note that the German central bank (the Bundesbank) has been particularly skeptical towards ECB QE. So, what happens if you look at researchers affiliated with the German central bank? Bundesbank researchers find much smaller effects of QE. In fact, Bundesbank researchers find an even smaller effect of QE on economic output than independent academics. Again, this indicates that the preferences of an institution seem to influence the conclusions of its researchers.

The paper presents additional analyses, such as looking more closely at the mechanisms at play, i.e. career concerns, involvement of management in research, and so on. Read the paper if you want to know more about this.

Conclusion

I think the paper by Brian Fabo, Martina Jancokova, Elisabeth Kempf, and Lubos Pastor is interesting also because it summarizes what the average effect of QE is, based on a large number of studies. Across more than 50 papers, the average maximum effect of QE on GDP and the price level is around 1.5%. This is useful information in itself.

Some papers are written by central bank researchers and some by independent academics. It seems reasonable to hypothesize that central bankers might have a tendency to view their own policies in a more favorable light. This is what Lubos and co-authors find. They report that the effect of having central bankers as authors of an analysis is almost as large as the average reported effect of QE on the price level itself. In private email correspondence, they also tell me that the average effect, estimated in academic papers, of QE on the price level is only 0.5%, i.e. very small, almost negligible. This is of course a controversial result. I predict it will generate intensive debate.

The fact that I discuss this paper here should not be taken to imply that I am skeptical towards central bank research in general. In fact, I am sure the quality of monetary policy decisions is improved when central bankers have access to the latest research. Also, I have no reason to believe that central bank research on other topics than monetary policy should be biased. But, when it comes to assessing their own actions, researchers in central banks might be subject to certain biases. It requires some guts to tell senior management that the trillions they have spent on quantitative easing probably has not been very effective. Instead, it might further your career if you paint a rosier picture. This is important to recognize.

I view the bottom line as follows: QE probably has some effect, but its effect is considerably smaller than we are told by central banks.

The Fed’s “Whatever it takes” moment. Or, how the Fed saved equity and credit markets

Facing a looming recession and financial market panics, the Fed intervened heavily in late-February/early-March, lowering the Fed Funds Rate to zero and expanding its balance sheet dramatically. In spite of this, markets kept on panicking. Then, suddenly, on March 23, everything changed. Stock markets started their rally. This was not because the Fed lowered the rate or expanded its asset purchases even further, nor because the economic data improved. What happened? The Fed made an announcement. Nothing else. It is a fascinating illustration of how expectations can change everything on financial markets.  

Much has been written about the massive interventions of the Fed during February and early March. In my previous post (link), I list Fed interventions as one of the reasons why the stock market is back to pre-crisis highs. In this post, I dig one step deeper and explain the fascinating story of how the Fed said something and thereby rescued markets.

Asset purchases and rate reductions did not save markets in February/March

Let us start by illustrating how the actual Fed interventions (interest rate changes and asset purchases) did not save markets in March. The Fed lowered the (lower range of the) Fed Funds Target Range to 1% from 1.5% on March 2 and then again to 0% less than two weeks later, on March 14. At the same time, it bought Treasuries and mortgage-backed securities to the tune of USD 600bn per week. These interventions succeeded in lowering yields on government and mortgage-backed bonds, but did not cheer up stock markets.

This graph shows how Treasury yields came down significantly in February/March, by basically 1.5%-point (from close to 2% to close to 0.5%; I show yields on 10-year Treasuries in this graph), as a result of reductions in the policy rate (the Fed Funds Rate) and asset purchases by the Fed.

SP500 and yield on 10-year Treasuries. Daily data.
Source: Fed St. Louis Database

The graph also shows that the SP500 continued falling throughout February/March. In other words, the massive interventions by the Fed in late-February/early-March (and these interventions really were massive – buying for USD 600 bn per week and lowering rates to zero is indeed a massive intervention) did not convince stock markets that the situation was under control. And, remember, these were not minor stock-market adjustments. It was the fastest bear market ever (link).

What turned the tide?

On March 23, the SP500 reached its low of 2237, a drop of 31% compared to its January 1 value. Since then, everything has been turned upside down and markets have been cheering, as the above graph makes clear.

What happened on March 23? The Fed had its finest hour. It did not do anything. It merely said something. A true “Whatever it takes” moment.

As you remember, a “Whatever it takes” moment refers to the July 26, 2012 speech by then ECB-president Mario Draghi (link). The speech was given at the peak of the Eurozone debt crisis. The debt crisis pushed yields on Italian and Spanish sovereign bonds to unsustainable levels. Italy was too big to fail, but also too big to save. The pressure on Italy was a pressure on the Eurozone construction. Mario Draghi explained the situation and said the by-now famous words:

“But there is another message I want to tell you. Within our mandate, the ECB is ready to do whatever it takes to preserve the euro. And believe me, it will be enough.”

Investors understood that the ECB would be ready to buy sovereign bonds to save the Euro. Markets calmed down. Italian and Spanish yields fell. And this – and this is the point here – without the ECB actually intervening, i.e. without the ECB buying Italian or Spanish bonds. The announcement that the ECB would intervene was enough to calm down investors.

Why is this relevant here? Because on March 23 the Fed sent out a press release, announcing that:

“…the Federal Reserve is using its full range of authorities to provide powerful support for the flow of credit to American families and businesses.”

And, then, as one of the new features announced the:

“Establishment of two facilities to support credit to large employers – the Primary Market Corporate Credit  Facility (PMCCF) for new bond and loan issuance and the Secondary Market Corporate Credit Facility (SMCCF) to provide liquidity for outstanding corporate bonds.”

This statement turned everything upside down. The important thing is that this turn of events happened without the Fed using any money at all. It was a “Whatever it takes” moment (though, perhaps not as Dirty-Harry dramatic as Mario Draghi’s “And believe me, it will be enough”): The Fed announced what they would do, investors believed the Fed, and markets started cheering. It is a prime example of how investor expectations influence financial markets.

What are the PMCCF and SMCCFs, why did the Fed announce them, and why were their effects so dramatic?

The lowering of the Fed Funds Rate and the purchasing of Treasuries succeeded in lowering yields on safe assets, such as Treasury bonds, as explained above, but did not lower yields on corporate bonds. In fact, during the turmoil in late-February/early-March, credit spreads (the spread between yields on corporate bonds and safe bonds) widened dramatically. When yields on corporate bonds rise, it becomes more expensive for corporations to finance their operations. And, when yields rise a lot, as in February/March, investors get nervous about the profitability and survival of firms.

This graphs shows how yields on both the least risky corporate bonds (Triple A) and speculative grade bonds (Single B) rose dramatically in relation to 3-month Treasury Bills, with yields on lower-rated bonds (Single-B) naturally rising more than yields on higher-rated (Triple-A) bonds.

Difference between yields on ICE BofA AAA US Corporate Index and 3-month Treasuries as well as the difference between yields on ICE BofA Single-B US High Yield Index and 3-month Treasuries. Daily data.
Source: Fed St. Louis Database

The important point in the picture is that the spreads continued rising throughout late-February/March, in spite of the intensive interventions described above, i.e. in spite of Fed purchases of government and mortgage bonds. The Fed was happy that safe yields fell, but was concerned that credit spreads kept on rising. Volatility in corporate bonds markets also rose (link), making the whole thing even worse.

More or less all firms saw their funding costs increase, even when there were differences across firms in different industries, with firms in the Mining, Oil, and Gas, Arts and Entertainment, and Hotel and Restaurant sector hit the hardest, and firms in Retail and Utilities sectors less affected (link). The Fed became nervous because higher funding costs for firms affect firms negatively, causing them to cut jobs, reduce investments, etc.

The Fed decided to act. It announced on March 23 that it would launch two new programs, PMCCF and SMCCFs. The PMCCF is the ’Primary Market Corporate Credit Facility’ and the SMCCF the ’Secondary Market Corporate Credit Facility’. Primary markets are where firms initially sell their newly issued bonds. The PMCCF should thus ease the issuance of newly issued corporate bonds, i.e. help firms raise funds. Secondary markets are where bonds are traded afterwards, i.e. the SMCCF should ease the trading (liquidity) of already existing corporate bonds.

The effect on equity and credit markets of the announcement of the PMCCF and SMCCFs was immediate and spectacular. Immediately after the announcement, credit spreads narrowed (see, e.g., link, link, and link).

Interestingly, stock markets reacted immediately, too. The stock market, thus, did not react to the lowering of the Fed Funds Rate and the extensive expansion of the Fed balance sheet in late-February/early-March, but reacted strongly to the announcement that the Fed would buy corporate bonds on March 23.

This (supercool, I think 🙂 ) graph shows developments on corporate bond and equity markets. The graph shows the spread between yields on AAA-rated corporate bonds and 3-month Treasury Bills and the stock market inverted, both normalized to one on January 1. The graph for the inverted stock market means that when the SP500 is at 1.45 on the y-axis on March 23, the stock price at January 1 was 45% higher than it was on March 23.

Difference between yields on ICE BofA AAA US Corporate Index and 3-month Treasuries (Left hand axis) as well as the SP500 inverted (Right hand axis). Both series normalized to one on January 2, 2020. Vertical line indicates March 23.
Source: Fed St. Louis Database

The parallel movements in equity and credit markets are striking. Equity and credit markets moved in parallel during January, when stock markets rose and credit spreads narrowed, in late-February/early-March when credit spreads widened dramatically and stock markets fell like a stone, as well as after March 23, when both credit spreads and stock markets improved spectacularly. Since then, stock markets have continued to rise and credit spreads have continued to narrow.

The correlation between the two series is an astonishing 0.93 for the January 2 through May 31 period.

The announcement effect

The Fed has experience with and a mandate to buy mortgage-backed securities and Treasuries. It had no such experience when it comes to purchases of corporate bonds and corporate bond ETFs. This means that the Fed could not start buying corporate bonds on March 23, it needed an institutional set-up. It created an SPV with capital injections from the Treasury and leverage from the New York Fed, it asked a financial firm (Blackrock) to help them purchase the bonds, etc. These things take time. The Fed only started buying corporate bond ETFs in mid-May and corporate bonds in mid-June.

Given that the Fed only started buying bonds and ETFs in May/June, the spectacular turnaround on March 23 really was due to the announcement only.

In slightly more detail, it played out as follows. The March 23 announcement primarily dealt with higher-rated corporate bonds (Investment Grade), and spreads narrowed immediately. On April 9, the Fed announced that they would expand “the size and scope of the Primary and Secondary Market Corporate Credit Facilities (PMCCF and SMCCF)”, buying bonds that were Investment Grade on March 22 but had been downgraded since then as well as corporate bond ETF. High-yield spreads tightened even more (link). Today, the Fed explains that the programs allow the Fed to buy “investment grade U.S. companies or certain U.S. companies that were investment grade as of March 22, 2020, and remain rated at least BB-/Ba3 rated at the time of purchase, as well as U.S.-listed exchange-traded funds whose investment objective is to provide broad exposure to the market for U.S. corporate bonds.”

The actual purchases began in May/June, several months after the announcement. Interestingly, the purchases themselves have had a modest impact only. First, the amounts of corporate bonds and corporate bond ETFs bought pale in comparison with the amounts of mortgage-backed securities (MBS) and Treasuries bought. The Fed has bought MBS and Treasuries to the tune of USD 3,000bn this year. It has “only” bought corporate bonds and ETFs for around USD 12bn. The amount used to buy corporate bonds thus corresponds to less than 0.5% of the amount used to buy MBS and Treasuries. This graphs shows the daily purchases. Using the latest figures from August 10, the Fed has basically stopped buying corporates.

Daily Fed purchases of corporate bonds and corporate bond ETFs. Daily data.
Source: Fed webpage. Thanks to Fabrice Tourre for collecting and sharing the data.

So, the Fed has not bought a lot of corporate debt. It has, however, the power to buy a lot. The PMCCF and SMCCF set-up is such that the Treasury has committed to make an USD 75bn investment in the SPV that buys the assets (USD 50bn toward the PMCCF and USD 25bn toward the SMCCF). The New York Fed then has the ability to level this up by a factor of ten, i.e. the Fed can buy corporate debt for up to USD 750bn. This is a sizeable fraction of the total US corporate bond market. This, that the Fed can potentially buy a lot, helps making the program credible and thus helps explaining its powerful impact.

Dilemmas

Given the success of the PMCCF and SMCCFs, commentators have started arguing that the Fed should be allowed to buy corporate bonds as part of its standard toolkit (link). This might be relevant, but purchases of corporate bonds by central banks raise a number of dilemmas:

  • Keeping zombie firms alive. By easing up stresses in corporate bond markets, the Fed calmed down markets. This was the intention of the PMCCF and SMCCF announcements and it worked. It eased access to funding for firms, and firms have raised a lot of cash as a consequence (link). It is positive that malfunctioning markets are stabilized, but if central bank intervention makes markets too cheerful it may allow firms that in principle should not receive funding to nevertheless get it. And, thereby, to keep firms alive for too long, and increase firm leverage too much. Basically, the fear is that programs such as the PMCCF and SMCCFs create too many zombie firms. My CBS colleague Fabrice Tourre and his co-author Nicolas Crouzet has an interesting paper that examines this (link). Fabrice and Nicolas find that when financial markets work perfectly (no disruptions), Fed intervention might be detrimental to economic growth. On the other hand, if markets are disrupted, Fed intervention might prevent a too large wave of liquidations. One thus needs to determine when markets are disrupted “enough” to rationalize interventions in credit markets. This is no straightforward task.
  • Distributional aspects. By buying some bonds but not others, the Fed exposes itself to the critique that it helps some firms at the advantage of other. To alleviate such criticism, the Fed has been very transparent and publishes a lot of information about the bonds it buys, the prices at which it buys the bonds, etc. (link). Nevertheless, it is easy to imagine that some firms at some point will start saying ‘Why did you buy the bonds of my competitor, but not my bonds’?
  • The Fed put. The more the Fed intervenes when troubles arise, the more investors get reassured that the Fed will also come to the rescue next time around. When the Fed saves markets, it is sometimes called the Fed exercises the “Fed Put”. The potential problem here is that if investors believe that the Fed will exercise the Fed put, investors will be tempted to take on even more risk. Those of us concerned about systemic risks get nervous.

So, in the end, the announcement of the PMCCF and SMCCFs was crucial during this crisis. There are, however, important dilemmas that need to be addressed when evaluating whether such programs should be part of the standard toolbox. I am not saying they should not. I am saying that one needs to be careful.

Conclusion

The Fed launched massive “traditional” interventions in late-February/early March, lowering the Fed Funds Rate and buying mortgage and government bonds. In spite of these very large interventions, equity and credit markets kept on tail spinning. When firms struggle, it hurts economic activity and employment. The Fed got nervous.

The Fed announced – and this is the whole point here; they only announced – that they would start buying corporate bonds. Markets turned upside down. Credit markets stabilized, credit spreads narrowed, corporate bond-market liquidity improved, and stock markets cheered. And, all these things without the Fed spending a single dime until several months after the fact. Even today, the Fed has spent very little (some might say that USD 12bn is a lot, but compared to asset purchases of USD 3,000bn, it pales).

It was a “Whatever it takes” moment. It illustrates how managing investor expectations can be crucial. Understanding this announcement is thus important for understanding the behavior of financial markets during this pandemic.

The weird stock market. Part II: Potential explanations

The behavior of the stock market during this recession raises three main questions: (i) why did stock markets fall so spectacularly during February/March, (ii) why did stocks rebound so spectacularly during April/May, and (iii) why is the stock market currently at its pre-crisis level? My previous post (link) presented the stylized facts and addressed stories that cannot explain the facts. In this post, I provide some potential explanations.

I my previous post (link), I mentioned that the rebound in the stock market since its low in March cannot be explained by (i) revisions to the economic outlook – in fact, economic forecasts were revised down during spring, (ii) a preference for just a few (FAANG) stocks, as the rebound has been international and broad based, even when large-cap stocks have done particularly well, (iii) a hypothesis that this recession is not particularly bad – in fact, at least in the short run, this recession is much worse than the 2008 recession, or (iv) the fact that markets also recovered after 2008 – markets always recover, i.e. this is not an explanation.

What can explain it, then? First, a warning: I do not want to raise expectations too much (sorry). I will not be able to offer a full explanation of these puzzles in an app. 1,900 words blog post. Also, as you will see, puzzles remain, even if we can explain some things. Finally, I have not seen a good explanation of these puzzling stylized facts (though Gavyn Davies’ piece in the FT is a good place to continue after reading this post; link). Hence, it would not be serious nor academic if I claimed that I could explain everything. What I can do, though, is to provide some hints at what might be relevant parts of an explanation.

Explanations that contain elements of truth

Earnings suffered more in 2008               

In the end, earnings and not economic activity (GDP) determine stock prices. In the long run, there is a strong relation between earnings and economic activity, which is why we focus on economic developments when we try to understand stock markets. The relation is not one-to-one, though, in particular in the short run. Hence, let us discuss earnings.

Earnings suffered dramatically in 2008. Often, we look at 12-month earnings, as quarterly earnings are volatile. 12-month reported earnings per share for the S&P500 came in at 6.86 in Q1 2009, which is app. 10% of 12-month reported earnings in Q1 2008. I.e., earnings fell by a mind-blowing 90% during the financial crisis of 2008.

Earnings fell in Q1 2020, too, but much less: From USD 139 per share in Q4 2019 to USD 116 (12 month reported earnings), i.e. by 17%. At the time of writing, around 90% of companies in the S&P500 have reported earnings for Q2 2020. Earnings (12-month reported) seem to be coming in at close to USD 96 per share, which is a drop of 17%, too, compared with Q1. Earnings have thus dropped by 17% during each of two consecutive quarters. But, and this is the main thing, these drops pale in comparison with autumn 2008. In autumn 2008, earnings dropped by 68% from Q3 to Q4 and then again by 54% from Q4 2008 to Q1 2009. This figure shows the quarterly percentage changes in 12-month reported earnings of the S&P500 in 2008 versus 2020. “0” on the x-axis is Q2 2020, respectively Q4 2008. The drop in earnings was just so much larger in 2008.

Quarterly changes in 12-month reported earnings per share, S&P500. “0” is Q2 2020, respectively Q4 2008. For Q3, 2020 and forward, these are expected earnings.
Source: https://www.spglobal.com

“1” on the x-axis in the figure is thus Q1 2009 and Q3 2020, i.e. firms’ expected earnings in Q3 2020 (and actual earnings in Q1 2009). If these expectations hold true, earnings are expected to do relatively fine going forward.

Earnings do fall from Q1 to Q2 this year. Earnings surprises matter, too, though, i.e. whether actual earnings are higher (or lower) than expected. At the time of writing, around 80% of companies have reported positive earnings surprises for Q2. This, that actual Q2 2020 earnings are better than expected, helps explaining why stock prices rise currently.

If investors primarily look at earnings, this helps explaining why stock prices fell considerably more in 2008 than they have done during this recession and why the stock-market recovery took longer in 2008. Earnings have simply not suffered so much this time around. For that reason, stock prices have not suffered so much this time around either.

Monetary and fiscal policies have been aggressive

Another factor that helps explaining why stock prices today are at pre-crisis levels, in spite of the severity of the recession, is that policy interventions have been aggressive. Immediately, at the onset of the crisis, monetary policy turned very expansionary. This can be illustrated by the weekly changes in the Fed balance sheet, as in this graph:

Weekly changes in the Fed balance sheet. USD mio.
Data source: Fed St. Louis Database

The figure plots the weekly changes in the Fed balance sheet due to purchases of financial assets, in millions of USD. The Fed balance sheet changes when the Fed buys/sells financial assets. During March this year, there were weeks when the Fed intervened to the tune of USD 600bn. This is an enormous amount of money. It is, by way of comparison, more than twice the weekly amounts spent during the financial crisis of 2008, the figure also shows.

At the same time, yields are very low. This forces investors to buy risky assets if they want some kind of return, supporting stocks.

In addition to very aggressive monetary policy, fiscal policy has also been very aggressive.

Why did stock prices fall during February/March?

OK, so earnings have not fallen as much as in 2008, earnings for Q2 2020 have surprised positively, monetary and fiscal policies have been very aggressive, central banks pump liquidity into the system, and interest rates are very low, implying that investors need to invest their money in risky assets if they want a positive expected return. These features help explaining why stock prices today are not much lower than in the beginning of the year. But, if this is the story, why did stock markets drop so dramatically in March?

Changes to expectations to earnings

Earnings have dropped in Q1 and Q2, even if considerably less than in 2008. Perhaps the stock-price drop in February/March was simply due to a downward revision in expected earnings. Alas, this is not the case. Landier & Thesmar (link) have an interesting analysis where they look at analysts’ expectations to the earnings of S&P500 companies during Q1 and Q2 2020. They find that analysts cut their expectations to earnings of S&P500 firms, but not so much that it can account for the drop in stock prices during February/March. In other words, the drop in stock prices in February/March cannot be explained by analysts reducing their earnings forecasts.

If the drop in expected earnings cannot account for the drop in stock prices in February/March, what happened then? Financial economists have a straightforward way of explaining this. Stock prices are discounted cash-flows. To understand stock price movements, we must understand cash-flow and discount rate movements. When the fall in expected earnings is not large enough to account for the drop in stock prices, discount rates must have increased in February/March. In other words, during February/March investors required a higher expected return if they should invest in stocks. This is also the conclusion in Landier & Thesmar and in Gormsen & Koijen (link). Basically, this is the conclusion in most empirical asset-pricing literature: stock prices move too much to be justified by movements in cash-flows. Discount rate variation is the reason, the literature concludes.

When stock prices move, and it cannot be because of cash-flow movements, it must be discount-rate movements. This is fine, but it is also somewhat tautological: If A = B + C, and A moves but B does not, then it must be because C moves.  A more difficult question is what makes discount rates move in the first place? I.e., what is the deeper economic explanation? Perhaps/probably risk aversion spiked, when investors in February suddenly realized the severity of the virus. Perhaps/probably some market participants faced funding constraints. And so on. In other words, I agree that discount rates most likely spiked during March, but I am not 100% sure why and I do not know if the increase in risk aversion is large enough to account for the increase in discount rates, and thus account for the drop in stock prices.

Lingering doubts

So, a potential reason why stock markets have done better than in 2008 is that earnings have not suffered as much as in 2008 and central banks have flooded the market with liquidity (and yields are at zero). The reason why we saw a massive fall in stock prices in February/March, and an unprecedented rebound during April/May/June, then probably is that discount rates increased during the early phase of the crisis, and then stabilized. This is a story that somehow makes sense.

Two things are still strange (to me at least), though.

First, if stock prices are at their right levels now, why did they have to fall so much in March? OK, because risk aversion increased and funding conditions tightened. But, if risk aversion increased in February because investors understood that a recession was looming, why would investors turn less risk averse three weeks later, when economic forecasts just kept on deteriorating? Probably the Fed and the Treasury (via monetary policy and fiscal policy) eliminated the spike in risk aversion via aggressive policy interventions, but it is still not 100% clear how to reconcile these features of the data.

Second, the behavior of earnings has been surprising. During recessions, earnings normally contract much more than economic activity. So, when economic activity in 2020 contracts much more than in, as an example, 2008, a reasonable hypothesis is that earnings in 2020 would contract much more than in 2008. This has not happened.

To some extent, this graph summarizes both the explanation of the relatively mild response of the stock market to this recession and the remaining puzzle:

Percentage changes in US real GDP and earnings of S^P 500 firms from Q3 to Q4 2008 and from Q1 to Q2 2020.

The figure shows how earnings dropped 68% from Q3 to Q4 2008 but seems to drop “only” 17% from Q1 to Q2 this year. This explains why the stock market has performed considerably better during this recession compared to its performance in 2008, for instance. On the other hand, the figure also shows that GDP dropped by 10% from Q1 to Q2 this year and “only” by 2% from Q3 to Q4 2008, i.e. the fall in economic activity is five times larger during this recession. It is puzzling that earnings drop so much less during this recession when economic activity contracts so much more.

So, the relatively modest contraction in earnings helps explaining the relatively fine performance of the stock market during this recession, but it is difficult to understand why earnings have behaved reasonably well. Perhaps this is because banks were suffering in 2008, but are able to help firms getting through this recession. Perhaps it is because firms have been able to adapt better this time. We do not really know. From an academic asset-pricing perspective, it is interesting that the stock market has behaved so differently this time around, compared to how it normally behaves. From an investor perspective, we can only hope that the stock market will keep on behaving in a different way than it usually does, as, otherwise, stock prices will fall going forward.

Conclusion

The facts that earnings have done better than in 2008 and that monetary policy has been very aggressive help us understand why the stock market has done well during this recession. Some puzzles still remain, though. For instance, why have earnings done so reasonably well, given the severity of the recession, and why did risk aversion increase so dramatically in February only to normalize few weeks later, in spite of the worsening recession. Perhaps the sensible answer is that this recession has been unusual in many dimensions, and some things will remain difficult to understand.

The weird stock market. Part I: Facts and wrong explanations

The behavior of the stock market during this recession has been perplexingly. To understand it, we need to answer three main questions: (i) why did stock markets fall so spectacularly during February/March, (ii) why did stocks rebound so spectacularly during April/May, and (iii) why is the stock market currently at its pre-crisis level? This blog post explains why these stylized facts are so difficult to understand jointly. In my next blog post (link), I will offer my interpretation of the evens.

Observing the stock market during this recession has been a like watching a roller coaster. The fall in late-February/early-March was the fastest bear market in history, the rebound has been historically fast, too, and – in spite of the worst recession ever – the stock market is currently back at pre-recession levels. Somebody leaving for the moon in January and returning now would not be able to see any trace in the stock market of the worst recession ever. This is in stark contrast to normal recessions. People interested in financial markets and the economy should be puzzled.

The problem – in a nutshell – is that if you claim that you understand why stocks are currently at pre-recession levels, then you face a challenge explaining why we needed to go through the fastest bear market ever during February/March. On the other hand, if you claim that the fast fall and rise in markets was just a short-lived technical blip (perhaps due to funding squeezes and fear), then you face a challenge explaining why that technical blip should result in a loss to the tune of USD 20 trillion (link). Finally, if you claim that you understand why markets fell so fast in February/March (perhaps because we were facing the worst recession ever – which, by the way, is a good explanation), then you face a challenge explaining why markets recovered so strongly in April/May/June, as the recovery cannot be explained by a brighter economic outlook. In fact, we have seen continuous downgrades of the economic outlook. E.g., the IMF WEO in June lowered its April forecasts for global economic output to –4.9%, from –3.0%. A huge downgrade. Strange that stocks recover in spite of this.

In this post, I describe the facts. I also review a couple of suggested, but wrong, explanations. In my next post (that I will publish next week), I will offer my view on what has been going on.

The facts

Here is an updated version of a graph that I have presented earlier. It shows the Danish, US, emerging market, and world stock markets (MSCI), normalized to one on January 1, 2020:

Stock markets since January 1, 2020. MSCI indices.
Data source: Thomson Reuter Datastream via Eikon.

Stock markets reached a temporary peak on February 19, after which they tanked dramatically. As mentioned here (link), it was the fastest US bear market in history.

The rebound has been equally spectacular. The US stock market is already back at its January 1 level. World and emerging markets are still a few percentages behind, but, basically, they have also recovered from the crisis.

In the US, it was the fastest bear market ever, as mentioned. In this graph, I document that it has also been one of the fastest recoveries. I calculate rolling 11-week gains/losses in the SP500 during the last 50 years.

S&P 500, 11-week percentage changes.
Data source: Fed St. Louis Database

As you can see from the encircled peak on the right-hand-side of the graph, from its bottom on March 23, the SP500 gained app. 45% until June 8. The graphs shows that no other 11-week period since 1970 has witnessed such a large gain (I cherry-pick the 11-week period, but the main message is that the rebound has been spectacular, and this message is robust).

Wrong explanations

You hear/read many explanations. Some of them are just not correct.

It all boils down to FAANG

Some claim that it is not the overall stock market (typically these people refer to the SP500) that has recovered. Instead, some argue, it is all due to FAANG (Facebook, Apple, Amazon, Netflix, and Alphabet’s Google). It is true that the behavior of FAANG has been impressive and has driven a significant part of the SP500’s recovery, but this is not the whole story. As you see from the first graph above, emerging markets have recovered, too. This is obviously not due to FAANG. Also, some markets, such as the Danish, have recovered even more spectacularly than the US market. Not FAANG either. Even US small-cap stock have recovered. Not fully, but significantly. The Wilshire US Small-Cap Index gained an astonishing 57% during the March 23 through June 6 period. True, small-cap stocks are still (early August) 12% below their January 1 level, but when small-cap stocks have rebounded spectacularly, too, it emphasizes that this is not all about all FAANG, even when FAANG stocks have done very well.

Markets also recovered after the 2008 financial crisis

This is a no-brainer: Markets always recover. The point here is that this rebound has just been amazingly strong. In this graph, I compare the behavior of the SP500 during autumn 2008 and the corona crisis. “0” in the figure is September 19, 2008, respectively February 21, 2020:

S&P 500, daily closing prices. Normalized to “1” 25 days before September 19, 2008, respectively February 21, 2020. September 19, 2008 and February 21, 2020 marked by vertical line.
Data source: Fed St. Louis Database

During the first couple of weeks of this spring’s crash, it looked like autumn 2008. The pain was short-lived, though. Stock markets fell for a couple of weeks and then rebounded. In 2008, stocks just continued falling.

It has taken a couple of months to recover this time around. After the financial crisis of 2008, it took more than two years. This figure shows that the US stock market was back at its summer-2008 level in early 2011 only, and the world stock market later still.

Stock markets since January 1, 2020. MSCI indices.
Data source: Thomson Reuter Datastream via Eikon.

This recession is not as bad as 2008

It should be obvious by now that this argument is very wrong indeed. I have argued myself that there is hope that, when we get a vaccine and economic activity picks up again, the economic rebound will be stronger than in 2008, the reason being that we entered this recession with a more “balanced” economy than in 2008 (in 2008, we had a housing bubble, a weak financial sector, etc.). Nevertheless, the short-run loss this time around is just so much larger than in 2008.

Last week, we got figures for GDP in the US and Europe. This graph compares the cumulative falls in real GDP in the Eurozone and the US in Q3+Q4 2008 with the cumulative drops during the first two quarters of this year.

Cumulative change in real GDP in the US and the Eurozone. “2008” is Q3+Q4 2008 and “2020” is Q1+Q2 2020.
Data source: Fed St. Louis Database

US real GDP fell by more than three percent in Q3 and Q4 2008, whereas Eurozone real GDP dropped by five percent in autumn 2008. It was a severe recession in 2008. Nevertheless, it pales compared to this recession. This time around, US GDP has fallen by more than ten percent already and Eurozone GDP is down by a staggering 15% during the first two quarters of 2020.

What about the length of the recovery? After 2008, it took around two years before economic activity had reached its pre-crisis level. This figure shows for, e.g., the Eurozone that it will last more or less equally long this time (expectations for 2020 and 2021 are from the EU Commission, Spring 2020).

Development in real GDP for the Eurozone as of Q2 2008 and as of Q4 2019. Developments after Q2 2020, i.e. after 2 quarters in 2020, are expectations from the EU commission.
Data source: EU commission.

The figure shows that Eurozone GDP was 5% lower in Q4 2008 compared to Q2 2008 whereas it is 15% down in Q2 2020 compared with Q4 2019. After eight quarters, GDP is basically back at its pre-crisis level. This means that the recovery is expected to be stronger, but this recession is just very deep.

In other words, this recession is much deeper – one can multiply the drop in 2008 by a factor of three or so to get the drop this time – and it will take several years before economic activity is back at pre-crisis levels. This means that the short-run cost to society is bigger this time around (a deeper fall stretched over the same period).

And, in spite of this very strong recession, stocks are doing fine. That is perplexing.

Above, I have presented stories that cannot explain what has been going on. In my next post (link), I will offer my view on this.

Expected returns

The Council for Return Expectations published its forecasts of expected returns last week. There is a wide dispersion across asset classes. Returns on “safe” assets (government bonds) are expected to be very low, even negative, on the short horizon, whereas emerging markets equities are expected to return close to ten percent per year. The Council also publishes forecasts for standard deviations, correlations, and inflation.

As one of my external activities, I chair the Council for Return Expectations (link). The Council estimates expected returns, standard deviations, and correlations on ten broad asset classes. The estimates are widely used in the Danish financial sector and public discussions. Danish pension funds use the estimates when they calculate pension projections for their customers and banks use them when they make projections for how outside-pension savings can be expected to develop. Newspapers also write about them, too.

I will argue that the Danish financial sector has been a front-runner in designing a way to make such independent return assumptions. I hope the set-up helps improving the credibility of the projections banks and pension funds make.

In this blog post, I present the latest estimates from the Council. As it is the first time I present these estimates on this blog, I start out describing why the Council was established, how the Council works, and the procedure we use in the Council to find expected returns. Afterwards, I present the forecasts.

Background: The Council for Return Expectations

The Council was established (under a slightly different name) two years ago. The background was as follows.

Danes have large pension savings. Together with the Netherlands, Denmark has the largest pension savings in the world relative to GDP (We have a paper describing some of the key features of the Danish pension system here; link). Historically, pension savings have been guaranteed, meaning that pension holders were guaranteed a minimum average annual return on their pension savings. Because of low interest rates, longer life-expectancies, etc., Danish pension funds have shifted into so-called market-based pension products during the last decade or so. In these products, there is no minimum guaranteed return. This allows the pension funds to invest more freely, hopefully enabling them to secure higher risk-adjusted returns. Basically, you go from a constrained to an unconstrained (or at least less constrained) maximization problem, which should lead to a better outcome. However, when a pension holder is not guaranteed a minimum return, the expected pension payouts during retirement will obviously become more uncertain. The assumed expected returns and risks on the different assets that pension funds invest in thus become even more important (compared to a guaranteed pension product) for an individual’s expected pension payments during retirement.

A couple of years ago, I was approached by the Danish pension industry and asked whether I would help them design a set-up that could be used to generate expected returns on pension funds’ investments. The result was the following: based on inputs from international financial institutions, present expected returns over the next ten years on ten different asset classes, as well as long-run (> 10 years) expectations on two assets classes (stocks and bonds), and update these forecasts annually. The pension industry judged that an independent committee should specify and regularly update the return expectations, in order to secure arms-length. They asked me if I would chair this committee. The Committee is now called the Council for Return Expectations

Last year, banks in Denmark asked whether we (The Council) would be able to expand the set of return assumptions. The background was – probably fair to say – a scandal in the largest bank in Denmark. Danske Bank had advised some of its customers to invest in an investment product that the bank itself expected would yield a lower return than a bank deposit. I.e., the bank had advised its customers to invest in a suboptimal product, given expected returns. It led to the resignation of the interim Danske Bank CEO (link). The Danish banking sector concluded that arms-length was needed in determining return expectations used for investments outside pension savings. They asked the Council if we could include return expectations for a shorter horizon (1-5 years, in addition to the 1-10 years horizon) and a more frequent update of the return assumptions (twice a year, instead of annually). These forecasts were published last week.

Procedure for determining expected returns

We provide forecasts for nominal returns on ten broad asset classes. Having done extensive research on the determinants of expected returns myself, I know that even within a precisely defined asset class, forecasters can disagree substantially on the outlook for the asset class. When designing the set-up, I thus suggested that return expectations should be based on an arms-length principle and take estimation uncertainty into account. We do this by basing our expectations on inputs from several international investment houses.

We (The Council) receive inputs from Blackrock, J.P. Morgan, Mercer, and State Street. We are truly thankful for their help.

Blackrock, J.P. Morgan, Mercer, and State Street provide their estimates of expected returns, standard deviations, and correlations on each of the ten asset classes for each of the different horizons to the Council. The ten asset classes for which we provide expected returns are:

  • Government and Mortgage Bonds
  • Investment-grade bonds
  • High-yield bonds
  • Emerging market sovereign bonds
  • Global equity (developed markets)
  • Emerging markets equity
  • Private equity
  • Infrastructure
  • Real estate
  • Hedge funds

The horizons are:

  • 1-5 years.
  • 6-10 years.
  • 1-10 years.

Returns are nominal. Estimates are for the arithmetic average annual return per year during the different horizons.

Returns are hedged into euros, i.e. are euro returns, except from emerging market equities and local currency emerging market debt that are unhedged (emerging market debt is 50%/50% local/hard currency). The Danish kroner is pegged to the euro, i.e. euro returns are basically also Danish Kroner returns. Returns are before fees, expect from the last four asset classes (private equity, Infrastructure, real estate, and hedge funds) that are after fees to the funds that manage these types of investments, but before the fees to the Danish pension/mutual fund that in turn invests in the private equity funds, hedge funds, etc.

In addition, we provide forecasts for Danish inflation based on inputs from a number of Danish forecasters (the Danish central bank, ministry of finance, etc.).

The Council consists of three people: Torben M. Andersen (professor of economics, University of Aarhus; link), Peter Engberg Jensen (former CEO of Nykredit and current chairman of Financial Stabilitet; link ), and myself as Chairman. Our job in the Council is to specify the relevant asset classes and their characteristics, choose methods used to weight the inputs together, come up with reasonable forecasts, communicate these to the public, etc.

Forecasts of expected returns

The following table presents our forecasts of expected annual returns for the next five years, the next ten years, and years six through ten:

Source: Council for Return Expectations.

Over the next five years, we expect “safe” investments (government bonds and Danish mortgage bonds) to return a negative 0.3% per year on a pre-fees, pre-taxes, and pre-inflation basis. This is a low return. On a net-of-fees, after-tax, and real-terms basis, it is even lower. In the Danish media, much was written about this number (-0.3%). For finance professors and professionals, the number is no big surprise given low interest rates, but for ordinary investors, the publication of numbers such as these from trustworthy sources helps communicating the message that it is difficult to earn a decent return these days without taking on risk.

On a more technical term, the return on this asset class is the return to a bond portfolio consisting of 50% Danish mortgage bonds (that are triple-A rated, as you probably know), 20% Danish government bonds, and 30% Euro government bonds, with a duration of five years.

There is a wide dispersion across the different asset classes. For instance, we expect emerging market equity to yield 9.5% per year. There is thus an almost ten-percentage point difference between the expected return on the safest asset class and the asset class yielding the highest expected return.

Notice also that we generally expect higher returns after five years. We expect higher interest rates, causing capital losses and thus low returns during the first five years, but then higher returns later on. This in itself helps raising expected returns on other asset classes.

Standard deviations

In the Council, we want to stress that estimates of expected returns are surrounded by uncertainty. To the finance professionals, this is obvious (though, one might sometimes have the impression that even professionals tend to forget this), but to the ordinary investor, this is even more important to emphasize. The ordinary investor might otherwise conclude that when expected returns on emerging markets is 9.5% per annum, but government bonds are expected to yield a negative 0.3% per annum, I better put all my money in emerging market equity. We want to stress that this is a risky strategy. This ambition has guided us when it comes to our estimation of standard deviations.

The expected returns we present are constructed as simple unweighted averages of the expected returns of Blackrock, J.P. Morgan, Mercer, and State Street, asset-class by asset-class. They also provide us with their expected standard deviations. We do not use the simple averages of these as our estimates of expected returns, though. Instead, we regress the standard deviations we receive from the investment houses on the expected returns received from the same investment houses. The standard deviations the Council presents are then the fitted values from this regression. The main objective we achieve by following this procedure is that we make sure that there is a clear relation between risks and returns.

The standard deviations surrounding the estimates of expected returns are here:

Source: Council for Return Expectations.

We expect a 3.5% standard deviation of government and mortgage bonds (over the next ten years) whereas we expect a 29.6% standard deviation surrounding the estimated returns to emerging markets equity. There is thus a clear relation between expected risks and returns. This clear relation also appears from this graph that plots risks on the ten asset classes against their returns (1-10 years horizon):

Source: Council for Return Expectations.

One might say that the price we pay from estimating standard deviations in this way is that some of the individual estimates of standard deviations might differ slightly from market consensus. As an example, our estimate of the standard deviation of global equity is around 14%. Market consensus probably is that this is a little higher, at 16%-17%. On the other hand, the advantage we obtain from proceeding in this way, as mentioned, is that we secure a clear relation between risk and returns.

Correlations

We also present correlations between expected returns. Here they are for the first ten years:

Source: Council for Return Expectations

Correlations are simple averages of the correlations we receive from Blackrock, J.P. Morgan, Mercer, and State Street, asset-class by asset-class. The main thing to notice probably is that we expect all correlations to be positive. This is bad news for investors, one might argue, but we saw this during the March turmoil. In March, bonds and stocks both lost in value, i.e. bonds did not hedge the risk of stocks. Our correlations reflect this.

The correlations and standard deviations are used by the pension funds to calculate confidence intervals surrounding the individual’s expected pension income. I want to stress that it is advanced – and good! – that pension holders in Denmark not only learn about their expected pension income but also the uncertainty surrounding this expectation.

Inflation

We expect the Danish rate of inflation to be 1.2% over the next five years and 1.4% over the next ten years.

All the numbers

All the numbers – means, standard deviations, correlations, for all the horizons, etc. – are available on the webpage of the Council (link). On this webpage, addition information about the Council can be found as well.

Conclusions

We live in low-interest times. Investors might be tempted to invest more risky in order to generate a decent return. But what are decent returns? And how much extra risk do investors incur when investing more risky? In most countries, you get one set of answers when you visit one pension fund/bank/financial advisor, but another set of answers when you visit a different pension fund/bank/financial advisor. This makes investors uncertain and do not help building trust between the financial sector and investors. The financial sector in Denmark has found a cool way of presenting independent/arms-length forecasts. I hope that other countries might be inspired from this way of addressing this important issue.

—————————————————————————-

PS. In my last post (link), I wrote about a new paper I have written on the financial stability in the Nordics. The journal where the paper is published is now available (link). It contains my own article and articles on the banking union, bail-in debts, household debts, and macroprudential policy, as well as a number of enlightening discussions.

How stable is the Nordic financial sector?

In 2008, banks were too fragile given the risks on and off their balance sheets. Many banks failed, others were rescued by governments/taxpayers. The societal costs were enormous. In a new publication, I evaluate the robustness of the Nordic financial sector today. I conclude that the Nordic financial sector is more robust than in 2008. This is important because we are currently going through a severe recession. If banks today were as weak as in 2008, this recession would have been even worse.

Once a year, Nordregio publishes the Nordic Economic Policy Review. The theme of this year’s publication is “Financial Regulation and Macroeconomic Stability in the Nordics”. Because of my experience from analyzing the financial crisis in Denmark in 2008 (I chaired the government-appointed committee that investigated the causes and consequences of the financial crisis in Denmark, the Rangvid-committee, link), I was asked to write the first paper in the publication (link), assessing the robustness of the Nordic financial sector today.

It is a policy paper. No fancy equations and regressions, but straight-to-the-point analyses and conclusions. The Review was published Tuesday, June 16. At its launch, I participated in a panel debate with, among others, two former deputy-governors at Riksbanken (the Swedish central bank), a former governor at Riksbanken, and myself. In this post, I review my main conclusions and relate them to this crisis.

An ironic incident
I start my paper emphasizing that it is difficult too foresee financial crises.

As an example, the first-page headline – typed in large bold letters – in the 2008 Financial Stability Report of Nationalbanken (the Danish central bank) was: ‘Robust Financial Sector in Denmark’. The Financial Stability Report analyzed the robustness of the Danish financial sector. It was published in May 2008, i.e. only a few months prior to the outbreak of the worst financial crisis since the 1930s. Half of Danish banks disappeared following the financial crisis of 2008.

Even if this headline probably still haunts Nationalbanken, Nationalbanken was not alone in not foreseeing the financial crisis. On the contrary. Financial crises are almost per definition unpredictable. If it was generally accepted that a financial crisis was in the making, action would surely be taken to prevent it. In this sense, we tend to become surprised each time.

The first version of my analysis was written in autumn last year. The final version was submitted in January this year. Right before the corona crisis.

I had not seen the corona crisis coming. In particular, I had not seen how severe it would be. Clearly, I was not alone either. Financial markets, for instance, had not seen it, and financial markets summarize the average views of all investors. Stock markets, credit spreads, stress indicators, etc. did not react before the crisis was in fact happening. So, was my assessment of the financial sector in January – that I present below – “robust” in light of the events that have happened? I will return to this in the end.

Why an important question?

If financial crises, and more generally stress in financial systems, are so hard to predict, why do we bother? There is one main reason: The societal costs associated with financial crises are enormous.

In the paper, I present new calculations of the societal losses resulting from financial crises in the Nordics. I do as follows. Societal losses are calculated as forgone economic activity due to financial crises. And this, forgone economic activity, I calculate by projecting economic activity, from the start of a financial crisis, by the growth rate of economic activity. The result is a path of economic activity in the hypothetical event that there had been no crisis. I contrast this with actual economic activity during the crisis. The difference is the cost of the financial crisis.

In the paper, I present results from such calculations for Denmark, Finland, Norway, and Sweden for the crises in the early 1990s and the crises in 2008. Let me present just one of the calculations here. This figure shows the result for Denmark for the 2008 crisis:

GDP and hypothetical non-crisis GDP for Denmark.
Data source : St. Louis FRED database.

The blue line shows developments in real per capita GDP (in USD to allow for a comparison across the Nordics) whereas the red line shows hypothetical GDP assuming no crisis. Before the crisis, real GDP per capita was developing steadily around the growth trend. The crisis changed this dramatically. GDP fell 5% during the crisis, itself a huge drop in GDP. On top of this, however, the recovery was slow. In 2018, ten years after the crisis, GDP had not recovered to the level that could have prevailed had there been no crisis. In fact, ten years after the crisis, the accumulated difference between hypothetical no-crisis GDP and actual GDP amounts to 91% of GDP in 2008. One year of GDP was lost due to the crisis.

I present these calculations for the other Nordic countries and other crises. The editors of the publication, Lars Calmfors and Peter Englund, in their introduction, summarize the results as follows: “Losses accumulate to a staggering one or two years of economic output”. This is why it is important to deal with financial crises.

In the paper, I describe the underlying causes behind the crises in the Nordics, as well as similarities and differences across countries. I skip this here. Read the paper to get it.

Risks assessment

As the next step, I venture into a kind of risk assessment.

Literature on financial crises concludes that even when it is difficult to foresee crises, some variables tend to be more informative about risks to financial stability than others are. I mainly discuss credit growth, house price growth, and household leverage. I conclude that lights are not flashing red.

Credit growth is the indicator to which most attention is typically paid. For instance, credit growth is the most important indicator when the counter-cyclical capital buffer is determined in the new Basle capital-regulation regime. Credit growth has been low during recent years.

House prices and household leverage are other important indicators. In the Nordics, they have been growing for almost three decades. House prices are currently high, as are levels of household debt.

At the same time, it is difficult to use an indicator that has been increasing for almost three decades as a predictor of the timing of a turnaround. The literature typically uses a three-year window for house-price growth when predicting crises. House-price growth has not been particularly strong during the recent couple of years.

So, house prices and household debt levels are high today. But twenty years ago, they were higher than they were thirty years ago. And, ten years ago, they were higher than they were twenty years ago. Today, they are higher than they were ten years ago. It is very difficult to say when the level is “too high”, in particular when interest rates have been fallen throughout the last thirty years, too (I return to this below). Had there been an explosion in house prices during the last couple of years, like prior to the 2008 crisis, it would have been a different matter. But this is not the case.

I conclude that house prices and levels of household debt are high but their developments are not particular useful to predict the timing of a crisis, and thus not supporting a conclusion that “a crisis is around the corner”

Robustness of the financial sector

Traditional indicators of risks to financial stability are not flashing red in the Nordics, but my introductory point was that crises tend to arise in unforeseen ways. The financial system should thus be robust to withstand unforeseen shocks. Prior to the 2008 financial crisis, the financial system was clearly not resilient enough. Banks “too large to fall” had to be rescued by governments and taxpayers. As a consequence, increasing the resilience of the financial system has been high on the agenda since the financial crisis.

Capital and liquidity requirements have been tightened, stress tests are conducted more systematically now, assuming even more severe stresses than assumed in stress tests conducted before 2008, restructuring and resolution regime have been introduced, macroprudential instruments employed, etc. I conclude that the financial system in the Nordics is more resilient today. This graph is just one way of illustrating this:

Capital ratios of selected large Nordic banks, before and after the 2008 financial crisis.
Data source: Eikon.

The graph shows capital ratios of selected large Nordic banks before (2006) and after (2018) the financial crisis. Prior to the financial crisis, large banks financed around 8-10% of their risk-weighted assets with equity. Today, the ratio is between 18-20%. In other words, capitalization of banks has been basically doubled since the financial crisis.

Causes for concern

Does all this (no lights flashing red and a more resilient financial sector) mean that there is no cause for concern? That would be going too far.

I am concerned by the current low-interest rate environment. Or, rather, I am concerned about the consequences if interest rates at some point start rising.

Today, interest rates are very low in an historical content. I use this graph to illustrate:

Monetary policy rates in the Nordics.
Data source: Datastream via Eikon.

The figure shows monetary policy rates in the Nordics since the early 1990s. Rates have been constantly falling, reaching negative territory in Denmark and Sweden (Sweden now back at zero again).

As mentioned above, house prices and debt levels have been increasing for app. 30 years in the Nordics, most likely driven by the fall in interest rates. Higher house prices and levels of debt might be justifiable when interest rates are low, but should interest rates rise, they might not be sustainable any more. Furthermore, asset prices are even more sensitive to the interest rate when interest rates are low. This means that it takes less of an interest rate increase to cause drops in assets prices (such as house prices) at low levels of the interest rate. Interest-rate risk is important to monitor, in my opinion.

On the other hand, if interest rates stay low, this might squeeze bank profits, if banks are unwilling to pass on low or negative interest rates to customers. This seems to be the case. In spite of negative policy rates since 2012 in Denmark, it was only in 2019 that Danish banks started charging negative deposit rates towards retail customers, and still only on large deposits (typically above EUR 100,000). An incomplete pass through of negative rates hurts bank profits, squeezing their resilience.

Finally, in the paper, I discuss briefly the possibility that the next crisis might arise in nontraditional banking areas, such as the emergence of credit extension by non-regulated financial institutions or cyber risks.

Discussion

The publication (Nordic Economic Policy Review) also includes great comments from two discussants of my paper: Anneli Tuomenin, CEO of the Finnish Financial Supervisory Authority and Peter Englund, Professor at the Stockholm School of Economics. Both largely agree with my assessment that the Nordic financial system is more robust today. Anneli is, though, perhaps even more worried than I am about the current low interest rate environment and outlines an additional number of non-traditional risks, such as climate risk and cyberrisk. Peter discusses whether banks really are that much better capitalized today (has capitalization gone up or risk weights down?) as well as additional risks associated with low interest rates. Read their comments. The main point is, I believe, that all of us share the view that we are particularly concerned about risks associated with low levels of interest rates.

The publication also contains additional interesting articles on the banking union, on macro prudential regulation, on monetary policy and household debt, on bail-in instruments, and other interesting topics. If you are interested in financial stability, the publication should be a must-read.

Lessons in light of the corona crisis

As mentioned, I wrote this paper in January 2020, right before the outbreak of the corona crisis. In the paper, I conclude as mentioned that the financial sector is more resilient today than in 2008. Now, we are in the midst of a recession that will cause a considerably larger fall in economic activity (on the short run) than during the financial crisis in autumn 2008. In autumn 2008, banks went belly-up and many had to be rescued. What has happened this time around? No bank has gone bankrupt, no banks have been saved, spreads are up on financial markets but not as much as in autumn 2008 (they were briefly in March, but came down quickly again). In autumn 2008, banks cut lending dramatically. Today, banks are helping customers getting through the crisis. I think this is important to emphasize, as you can still find people who argue that banks are not much more robust than they were in 2008. I judge that they are.

I am not saying that all banks will survive this, and I am not saying that we will not face any threats to financial stability. But I am saying that, given the fact that this recession is even deeper (on the short run) than the 2008 recession, and the fact that we are not witnessing the problems we did in 2008, this indicates – at least until now – that the financial system is more resilient today. Of course, should the situation turn to the worse, e.g. with a new wave of the virus, things will look different.

Webinar

The paper was presented at a webinar on June 16, 2020. The presentation was followed by a panel discussion on “Will the corona crisis also trigger a financial crisis?”. Panelists included Karolina Ekholm, Stockholm University and former deputy-governor at Riksbanken, Lars Heikensten, the Nobel Foundation and former governor of Riksbanken, Kerstin Hessius, CEO of the Third Swedish National Pension Fund and former deputy-governor at Riksbanken, and myself. The panel was moderated by Peter Englund. It was an interesting debate.

Interventions that pay off. And, interventions that do not

In many European countries, lock-downs have been comprehensive. Now that we have more data, it is time to ask the question whether all types of interventions are equally effective. New research indicates that restricting large gatherings and public events is effective. Restricting international and internal travel is not.

Given the fact that the number of new cases continues to fall in Europe, even when more and more is opened up (link), I cannot help wondering whether all types of lock-downs are equally effective. It must so that some measures are more effective than others are. And, if some measures are not effective, they should be applied with greater caution if we will face a second wave of the Covid-19 virus.

It is understandable that lock-downs were comprehensive early on, as data and studies on the effectiveness of different non-pharmaceutical interventions were lacking. Now we have more data. A new study is interesting in this regard.

Nikos Askitas, Konstantinos Tatsiramos, and Bertrand Verheyden (link) use data from 135 countries to study the effectiveness of different types of non-pharmaceutical interventions. By studying such comprehensive data, Askitas et al. are able to exploit variation in the timing, type, and effectiveness of interventions within and across countries in order to identify the effects of different types of interventions. The data also allow the authors to control for the contemporaneous influence of multiple non-pharmaceutical interventions, i.e. they can do their best to isolate the effect of each type of intervention.

Askitas et al. classify interventions into eight different categories: (i) international travel controls, (ii) public transport closures, (iii) cancelation of public events, (iv) restrictions on private gatherings, (v) school closures, (vi) workplace closures, (vii) stay-at-home requirements, and (viii) internal mobility restrictions (across cities and regions). Their main results are here:

Source : Askitas, Tatsiramos, and Verheyden (2020)

The figure reveals how the number of daily new infections develop prior and subsequent to the introduction of different types of interventions. The figure shows that the most effective interventions are cancellations of public events and restrictions on gatherings. For instance, 35 days after the introduction of interventions that cancel public events, the number of new infections has fallen by 25%, compared to the number of new infections at the event time, i.e. at the time of introducing the intervention, on average across the 135 countries in the study. The figure also shows that the effect becomes significant only after app. 20 days, i.e. it takes time before the interventions work.

On the other hand, international travel controls, public transport closure, and restrictions on internal (within country) movements seem to have basically no effect on the number of new cases. In fact, these types of interventions do not seem to have any statistically significant effect on the number of new infections.

School closures seem to have some effect, although it is barely significant.

Other authors have studied the effectiveness of non-pharmaceutical interventions, even when these studies do not look as carefully as the Askitas et al. study on the effectiveness of different types of interventions. Robert Barro (link), for instance, studies data from the 1918 Spanish flu, exploiting variation in the intensity and duration of interventions across US state. Barro finds only little effect of interventions. Barro also reports, however, that this no-effect result is most likely due to the short duration of the interventions implemented during the Spanish flu. Most of the interventions back then had a duration of around one month. The figure above shows that interventions only start being really effective after one month or so. In this light, it is perhaps not surprising that Barro finds only little effect of non-pharmaceutical interventions.

Other studies, for instance link and link, also report a positive effect of non-pharmaceutical interventions but do not estimate as clearly the effect of different types of interventions.

So, the conclusion, at least to me, is that:

  1. Non-pharmaceutical interventions can be effective, in particular if their duration is not too short.
  2. Some interventions are more useful than others are.
  3. In particular, restrictions of large gatherings and public events seem to be rather effective.
  4. Restricting international travel and public transportation does not seem to be effective.

Hopefully, results like these will be read carefully by politicians. If we will face a second wave of Covid-19, activities that do not influence how the virus spreads should not be shut down. On the other hand, we should be fast in implementing interventions that are effective, and their duration should not be too short. In particular, there is no need to uncritically just shut down “everything”. It is perhaps understandable that this was the approach taken in many European countries in March when evidence on the effectiveness of different types of interventions was missing. Today, we know more. Research indicates that it pays off to restrict gatherings of large groups of people, while restricting international travel and public transportation does not seem to pay off.

The price of a lockdown

Has the lockdown been too strict and have we paid too high a price (in terms of forgone economic activity) as a result? Evidence from Denmark and Sweden helps answering this important question.

The fact that the number of deaths associated with Covid-19 falls in many European countries is unbelievably good news. As a result, European economies are opening up again. This is also good news. What is perhaps somewhat surprising, though, is that the number of deaths and new cases has not started climbing after opening up. This raises the question whether too many sectors were shut down and we, consequently, paid too high a price to contain the virus. Evidence from Denmark and Sweden suggests that the price was probably not as high as one might intuitively have expected.

Covid-19 in Denmark

The Danish prime minister announced the lockdown of Denmark on March 11. Death numbers kept climbing until early April, but has fallen steadily since then, as this figure shows

Number of new deaths in Denmark. Rolling 5-days average. Vertical lines indicate April 15 (primary schools opened), April 20 (dentists etc. opened), May 11 (retail shops opened), and May 18 (restaurants opened).
Data source: Statens Serum Institute. https://en.ssi.dk/

The fact that Covid-19 seems to be under control in Denmark is obviously very good news.

The lockdown has been costly, though. Jobs have been lost, firms have closed down, incomes have fallen, etc. Public debt levels will increase significantly.

Denmark started opening up again in mid-April. On April 14, private corporations that had been asked to encourage people to work from home were now encouraged to let workers return to their offices, subject to suitable safety measures. On April 15, kindergartens and lower primary schools opened up, and on April 20, hairdressers, dentists, etc. were allowed to open. If the lockdown helped contain the virus, one would have expected to see a rise in deaths and new cases after easing the lockdown. This has not happened.

On May 11, retail shops opened. On May 18, restaurants. Again, if the lockdown was the primary reason why the virus stopped spreading, we should have seen the number of new cases increase after opening up. This we have (luckily) not.

Given that the number of deaths and new cases has been constantly falling since opening up, one may wonder whether the lockdown was too strict. It seems that voluntary social distancing, avoiding large crowds, and the washing of hands have been perhaps even more important. Doctors and scientists wonder whether the virus has mutated and turned less infectious, or warmer weather in April and May has caused a slowdown in new cases, but if I understand what they are saying, they believe that hand washing (and related precautions) is a main reason why numbers fall.

If this is true, we need to ask the question whether it was necessary to keep children and workers at home and close down as much as we did. Whether the economic costs have been too high. This is a sensitive and difficult question, though, because it requires that we evaluate how much economic activity would have fallen had we not shut down as much as we did. Surely, people would have cut consumption anyhow, as people would have been afraid of the virus itself and its consequences. So, what was the extra price we paid because of the lockdown?

Evidence from a clever study

A clever study (Link) by University of Copenhagen researchers Asger Andersen, Emil Hansen, Niels Johannesen, and Adam Sheridan helps us answering this question.

Asger, Emil, Niels, and Adam compare spending by Danes and Swedes. Denmark and Sweden are similar along many dimensions, but have followed different strategies when it comes to Covid-19. Denmark enforced a strict lockdown, as mentioned, whereas Sweden has kept primary schools, restaurants, bars, etc. open (though did close universities and encouraged working from home). If the lockdown was the main cause of the recession, economic activity should have suffered more in Denmark than in Sweden. Andersen et al. use individual bank-account data from a large Danish bank that operates in both Denmark and Sweden (Danske Bank) to address this question. They track spending of 860,000 individuals before and during the lockdown, studying their use of cards, cash withdrawals, mobile wallets, etc. They compare spending of Danes and Sweden from mid-March (when Denmark shut down) through early April. This figure contains their most important result:

Data source: Andersen, Hansen, Johannesen, and Sheridan (2020).

The figure shows that spending of Danes fell 29% compared to a hypothetical spending path without the Covid-19 pandemic. A 29% contraction is enormous. But, and this is the main point, spending by Swedes fell by almost as much, 25%. This means that Danes cut spending by “only” 4%-points more than Swedes. Swedes could continue visiting restaurants, schools, etc., but Swedes cut spending dramatically nevertheless. The causal effect on spending, resulting from the extra lockdown in Denmark, is 4%-points and considerably smaller than the effect of the virus itself.

One may debate whether 4% is a small or a large number. In normal times, we would say that a 4% drop in aggregate spending is large. The main cause of this recession, at least according to this study, was not the lockdown, though, but the virus. In other words, even if economies had not been shut down as much as they were, the contraction in economic activity would still have been very large.

The broader picture

The paper by Andersen et al. weighs in on an important debate about the underlying cause of this recession. Is it due to a reduction in supply or demand? Their results indicate that the demand effect is the important one. Spending drops even when supply is less constrained.

But the strict lockdown in Denmark did cause an extra 4%-point drop in spending. Was it worth it? Nobody knows, but research from the MIT (Link) on the 1918 pandemic indicates that areas in the US that took more drastic measures early on, in terms of social-distancing measures and other public health interventions, had stronger recoveries after the measures were eased. There is debate how strong this result is, though (Link).

If all this is true, i.e. that the lockdown was not the main cause of the recession, but the main cause was the virus itself and the fear it created, and that the recovery is stronger afterwards in countries that imposed stronger interventions, we should see forward-looking measures indicating this. Early data from the Danish and Swedish stock market seem to point in this direction, even if the evidence is admittedly very tentative and a fuller analysis is obviously needed. This figure shows the Danish and the Swedish stock markets surrounding the date of the lockdown.

Danish (OMX Copenhagen 25) and Swedish (OMX Stockholm 30) stock market indices. Vertical line indicates March 12, when Denmark was shut down.
Data source: www.nasdaqomxnordic.com

I have normalized the two stock-market indices to “1” on March 12, the day after the lockdown was announced in Denmark. The Danish and the Swedish stock markets followed more or less similar paths during March, but, since then, the Danish stock market has gained 29% whereas the Swedish market has gained “only” 20%. Again, one needs a more detailed analysis (the Danish stock market is heavily influenced by pharmaceuticals and other defensive stocks that do relatively well in downturns), but this early evidence indicates that stock market investors are more positive on the Danish stock market.

Furthermore, the number of deaths is currently much lower in Denmark than in Sweden (also on a per capita basis), which is of course also something to remember:

Denmark and Sweden vs. other countries

The idea of using credit-card and other payment-type information is great and has been used by a number of economists to obtain real-time information on the economic impact of Covid-19. Microdata indicate that consumer spending in France fell by 50% as a result of the virus (Link), in Spain by 50% (Link), and in the US by something like 50%, too (Link). It thus seems that the contraction in spending is somewhat smaller in Denmark and Sweden compared to other countries. The comments above should be viewed in this light.

In any case, a 30% or 50% drop in spending is a huge drop. But, a 50% drop in spending does not mean that overall economic activity drops by 50%. Overall economic activity includes public spending, exports/imports, etc. So, credit card transactions provide useful information about consumer spending, but cannot stand alone when judging overall economic activity.

Conclusion

At some stage, we need to evaluate carefully whether too much was shut down. Not to blame somebody, but to learn. It is important that we analyze thoroughly whether it is better to be very aggressive early on in a pandemic or whether less aggressive measures are enough. Currently, it seems that we shut down a little too much (and we are now too slow in opening up e.g. borders again), but that the price of doing so was not as high as one might have expected intuitively.

Zero-probability events that happened

Some of the events we have seen during this crisis are so unlikely that we need tricks to calculate their probabilities. This is a journey into the world of very small probabilities and very large numbers. I admit it is somewhat surreal. But it is important as these small-probability events can have long-lasting effects on economic activity.

The defining characteristic of this recession is the speed with which markets and economies started freefalling. From one day to the next, economies were shut down. Economic activity came to a complete sudden stop in some sectors. In different blogposts, I have used words such as “unprecedented”, “completely crazy”, “horrifying”, “tremendous”, etc. to describe what happened.

Using words like these is adequate, I think (I admit I was shocked by what happened), but I have come to a point where I want to be more precise. As economists, we want to put numbers on things. I want to calculate the probability of the events. This turned out to be more complicated than I had first imagined.

The probability of a 121-sigma event

On Monday, April 20, the oil price (West Texas Intermediate) closed at USD -38 for a barrel of oil. I describe this here (link). This was a drop of 306% relative to the closing price the previous trading day. Based on daily data since 1983, the average percentage change in the oil price is 0.03% and the standard deviation 2.5%. I called the event “completely crazy”, which I still believe is a fair description. Now, I want to know “how crazy” it was. What was the probability that it would happen?

The happy-go-lucky Professor of Finance, i.e. me, opens his Excel spreadsheet and enters the command used to find the probability that an event of -306 happens when data are normally distributed with mean 0.03 and standard deviation 2.5.

I press “Enter”. Excel returns “0”. The Professor of Finance thinks this look strange and increases the number of decimal places. Excel just shows “0.000….”. According to Excel, this is a zero-probability event. It could not happen.

But it did happen. I start remembering the statistic classes many years ago, and recall the discussions on low-probability events.

Luckily, I have helpful colleagues. I write CBS Professor of Statistics Søren Feodor Nielsen. He tells me that the probability I am looking for is below the “machine precision”. He also tells me that the probability itself might be extreme, but the logarithm to the probability is probably not. We calculate the log-probability (using a different statistical package than Excel). It is -7495.137, i.e. the likelihood is exp(-7495.137). This is, unfortunately, such a low number that a computer cannot calculate this either.

We then calculate the log base 10 probability. Now we have the result. The probability is 0.799 x 10^(-3255).

This is a zero followed by 3255 zeroes, and then 799. Practically zero, but not exactly zero. The event happened, but it is extremely unlikely. We can call it “completely crazy”, but we can also say that the probability is 0.799 x 10^(-3255).

There is another way to illustrate how unlikely this event was. When the probability is 0.799 x 10^(-3255), we should see this event happen every 1.2516 x 10^(3255) days. I.e., a number with 3255 numbers before the decimal place. I simply do not know what such a number is called (link). This is where it starts getting surreal.

How precise is this? We have calculated it, but Søren tells me we should not put too much faith in its precision. Tail-probabilities in the normal distribution are calculated by numerical approximations, and it is questionable how precise they are when we are that far out in tail. Admittedly, the question is of course also how important it is that it is precise that far out in the tail. The probability is unbelievably small. That is the main thing. Exactly how small is perhaps not that important.

S&P 500

This graph shows daily percentage changes in the S&P 500 throughout the last 50 years.

S&P 500. Daily data. May 1970 – May 2020.
Data source: Thomson Reuter Datastream via Eikon.

On March 13, the S&P 500 fell 12%. The average daily percentage change in the S&P 500 (calculated up until March 12) is 0.03% and the standard deviation is 1.05%.

The likelihood that we will see a 12% drop in the S&P 500 on a daily basis, given the behavior of the S&P 500 during the last 50 years, is very small, but not so small that it cannot be calculated in Excel. Excel says it is 1.0827 x 10^(-29).

These are daily data. If we assume 250 trading days per year, we should see this event happen every 3.69 x  10^(27) = 3,694,454,429,465,560,000,000,000,000 year. I.e., every 3.694 octillion year.

This is also somewhat surreal.

There are many caveats. We are again so far out in the tail that we should not put too much emphasis on the exact number of decimals and the exact number after all the zeroes. Also, all these calculations are based on the assumption of normally distributed data. Given that we saw a 21% fall in the S&P 500 on Black Monday, October 19, 1987, we have seen two very unlikely crashes within the last three decades. According to these calculations, they should happen much more seldom. We use the normal distribution, but data might follow a different distribution. For most practical purposes, however, it is not necessary to know the exact distribution of these extreme events. The important thing to know is that the likelihood is very small.

Other numbers

In this post (link), I describe how the initial jobless claims soared, confidence indicators fell, GDP dropped, etc. As an example, the likelihood that we should see almost 7 million people filing for jobless claims on March 28, given the complete previous history of jobless weekly claims, is 0.97 x 10^(-841). This is again such a small probability that Excel cannot even calculate it.

Economic effects

There is a broader point to these discussions. Events like those described above might have long-lasting consequences for economic activity.

When events are unlikely, we assign low probability to their occurrence. One hypothesis is that we keep on assigning low probabilities to their occurrence. In this case, future investment decisions of firms and future consumption decisions of households are unaltered by the events we have been going through. It was a temporary shock. It has very low probability. It will not influence how we form expectations going forward.

An alternative hypothesis is that we have become so scared by the event that it will haunt us for years to come. We update our beliefs disproportionally much.

Which of the two scenarios play out is important for the recovery from this recession. Will we start consuming when economies open up or will we hold back because we have become scared?

Kozlowski, Veldkamp and Venkateswaran (2020) have an important paper on this (link). Their main point is that events like those studied above, i.e. events that are very unlikely but have large impacts, will have persistent effects on beliefs. They write that “tail events trigger larger belief revisions” and that “because it will take many more observations of non-tail events to convince someone that the tail event really is unlikely, changes in tail beliefs are particularly persistent”. This is an important insight. If this is how we form expectations, it will slow down the recovery from this recession. Kozlowski et al. also show, however, how government interventions can reduce the effect, but not eliminate it.

Conclusion

These weeks we are seeing infection rates declining in many countries and economies opening up again. These are very good news. Let us hope we can start returning to some kind of normality. It will take time before we are back to where we came from, though. Some things might even have changed for good. One thing is for sure: we will not get back as fast as economies and markets fell in March. What happened was very unlikely. But it did happen.

Is the stock market too optimistic? Or, is the market always right?

The rebound in stock markets has been spectacular. One may wonder whether it is sustainable.

What does it actually mean that the stock market is “too optimistic”? Doesn’t the stock price always reflect the average expectation of all investors? True, and, in this sense, the market cannot be “too optimistic”. On the one hand, the market is always right.

On the other hand, the legendary economist Paul Samuelson famously noticed that the “stock market has predicted nine of the last five recessions”. Stock markets also appear “excessively volatile”, as Shiller showed already in 1981. I.e., the stock market might always be right, but perhaps we are going through a period of excessive volatility where markets were overly pessimistic in March and are overly optimistic now.

In any case, if the stock market is always right, so is the futures market. Investors in the futures market expect huge drops in dividends over the next year. If stock prices behave as they usually do during recessions, stock prices should drop even more, that is dropping even more than dividends.

Market developments

Before we start this attempt to explain things and look ahead, let us review the facts. This graph shows developments in the Danish, the US, the world (MSCI All countries), and the emerging stock markets since the start of the year, all MSCI indices (Danish index is in Danish kroner, the rest in USD):

Stock markets since January 1, 2020. MSCI indices.
Data source: Theomson Reuter Datastream via Eikon.

Stock markets did well in the beginning of the year, fell dramatically from mid-February through mid-March, and have rebounded spectacularly since then. Today, the Danish stock market is above its January 1 value, the US market is 10% down, the world market is 15% down, and emerging markets are 20% down.

What could explain current stock prices?

Stock prices are by definition discounted future cash-flows. That is easy to say. The difficult part is to find the expected future cash-flows and the appropriate discount rate.

Let us start with a positive view on markets and see what is needed to support this.

To begin with, let us keep the discount rate fixed. Let us also, as a starter, assume that valuation ratios are constant. For instance, let us assume that stock prices closely follow developments in real economic activity. Under these assumptions, we would, largely, be able to understand recent market movements. These assumptions would also support a rosy view of the future. Afterwards, we discuss if this is a likely scenario.

Forecasts for economic activity in light of the corona crisis start coming in. Late April, we got the WEO from the IMF and forecasts for the US from the CBO, and this week we got forecasts for Europe from the EU-commission. This figure shows the expected path of real GDP in the US and EU:

Expected developments in EU and US real GDP. The CBO does not provide quarterly growth rates for 2021, so I assume linear growth during 2021, accumulating to the CBO forecast of 2.8% for 2021 .
Data source: EU commission and CBO.

This recession is enormous. GDP is expected to fall by 15% in Q2 this year, compared to late 2019. Luckily, economies are expected to rebound sharply after this quarter, too. The EU commission expects a strong recovery in the EU in 2021, with 6.1% growth in 2021. The CBO expects 2.8% growth in 2021 in the US. Whether this difference is realistic, I do not want to discuss here.

The economy drops by 15% in Q2. If stock prices follow GDP, as we assume for now, stock prices should have dropped by 15% in Q2. This is almost spot-on.

This theory would also predict that we have a great year in front of us. Economic activity should improve by something like 15-20% until late 2021. If stock prices follow, stock prices should also increase with something like 15-20% from here.

What about the huge drop in March? This we can also explain, I think. Basically, it had little to do with underlying economic fundamentals. Some leveraged hedge funds got squeezed, they received margin calls, they dumped everything to raise cash, there was panic, spreads widened, and markets feared a replay of 2008 (I comment on it here and John Cochrane has a nice and more detailed explanation here). Stress on markets almost reached 2008 levels, as, for instance, the St. Louis Fed Financial Stress index indicates:

St Louis Fed Financial Stress Index.
Data source: St. Louis Fed.

Central banks intervened and provided liquidity. Markets calmed down. The stress index is still somewhat elevated, but much lower than in March. Markets started looking at economic fundamentals again. With lots of liquidity around, and low yields, investors bought stocks. This was a temporary crash.

The intermediate conclusion is that if stock prices follow GDP, stock prices should drop by something like 15% during Q2, only to rebound afterwards. And, as of today, stock prices have in fact fallen by something like 15% since the start of the year, with a bumper on the road in March. If we stop here, everything would be fine and we should expect substantial stock market gains going forward. Under these assumptions, markets are expected to rebound by something like 15-20% until late 2021.

Cash-flows do not always follow economic activity

Unfortunately, we cannot stop here. We need to go back to the definition of stock prices: Discounted cash-flows. I truly believe that cash-flows relate to economic activity in the long run (I have research demonstrating this), but I also believe that there are temporary business-cycle deviations between cash-flows and economic activity. And these deviations can be substantial.

We have expectations to economic activity from IMF, CBO, EU, and so on, but how do we find expectations to cash-flows? Niels Joachim Gormsen from the University of Chicago, a smart former Ph.D. student at Copenhagen Business School, has, together with Chicago Professor Ralph Koijen, developed a method that can be used to back out expected changes in dividends from dividend futures. They also estimate a relation between dividend growth and GDP growth, such that they can back out expected GDP growth, too.

Their latest estimates are from April 20. Compared to January 1, 2020, Gormsen and Koijen estimate that US GDP will be 3.8% lower over the next year (precisely, Niels tells me, they compare expected growth from April 1, 2020 through April 1, 2021, to expected growth from January 1, 2020 through January 1, 2021). This is not far from what the CBO expects (the figure above shows that GDP will be something like 5% lower ultimo 2020 compared to primo 2020).

The scary thing is that Gormsen and Koijen show that markets expect dividends to be 18% lower over the next 12 months, i.e. drop by 13%-points more than GDP. An 18% drop in dividends is a very large drop in historical terms.

This graph shows how investors update their expectations to future dividend growth rates during the corona crisis:

Data source: Gormsen and Koijen (2020).

For the EU, Gormsen and Koijen expect GDP, respectively dividends, to drop by 6.3%, respectively 28%. The market is always right, right?

Unfortunately, we cannot stop here either.

Stock prices react excessively to cash-flows during recessions

We need to talk about the last part of the definition of stock prices: discount rates. Research shows that discount rates move counter-cyclically, increasing in bad times and dropping in good. From our definition of stock prices, this means that stock prices should fall by more than dividends in bad times, if discount rates increase in bad times.

Tim Kroencke from the Univeristy of Neuchatel has an interesting paper that studies how stock prices move in relation to dividends during recessions. The central graph in the paper is this one:

Data source: Kroencke (2020)

The graphs shows how dividends (and earnings) fall during recessions and how stock prices fall even more (there are two stock price indices. Read Tim’s paper to get the explanation for the difference. The point is that stock prices fall more than dividends.)

The figure shows that US dividends drop by 10% and US stock prices drop by ten percentage-points more, i.e. 20%, on average during US recessions. It also shows that stock prices do not drop ahead of the drop in dividends, but alongside. This is not good news.

Gormsen and Koijen’s estimates say that we should expect dividends to drop by 18% for the US and 28% for the EU over the next year. Kroencke says that on average stock prices drop by 10%-points more. This means that stock prices should drop by close to 30% in the US. If we use the 10% drop in the price-dividend ratio for EU data, EU stock prices should drop by close to 40%. Today, stock prices are down 15% globally.

Conclusion

We might hope that this recession plays out in a different way than recessions usually do. We might also hope that stock markets react in a different way than they usually do. We might hope that everybody starts spending when economies open up, companies start producing, and earnings and dividends do not suffer. In this case, the drop in dividends might not be that large. And, we might hope that stock markets do not fall by more than dividends, in contrast to what they normally do. This could for instance be because central banks are marginal investors in many markets now, driving asset prices away from what they would have been if prices were solely determined by private market participants. This is what markets hope. And, of course, this might turn out to be the case.

We might also fear, however, that companies will start facing problems given the severity of the crisis, i.e. this turns from a liquidity crisis to a solvency crisis. Should this happen, earnings and dividends will drop. The market is always right. The futures market expects dividends to drop by something like 20% for the US and 30% for Europe. If stock prices react as they normally do during recessions, stock prices should drop even more. The 15% drop since early 2020 seems small in this light.

———————————————————————————-

PS. Did you by the way notice the German Constitutional Court ruling this week? Given last week’s post on the Italian situation, I might just inform you that the Italian yield spread to Germany widened as a consequence. Not a lot – 15 basis points or so – but in the wrong direction.