Predictable financial crises
Robin Greenwood, Samuel G. Hanson, Andrei Shleifer, Jakob Ahm Sørensen 15 July 2020
There is a long-standing debate on whether financial crises can be predicted. This column draws on a chronology of past financial crises and data on credit and asset prices for a panel of 42 countries between 1950-2016 and finds that if there is a large credit expansion with an asset price boom, then financial crises are highly predictable. These results are used to motivate a simple indicator that identifies periods of potential credit-market overheating. The indicator is shown to predict past crises in advance, suggesting that policymakers have time to act and take prophylactic policy interventions.
How predictable are financial crises? An important line of thought postulates that they are largely unpredictable. Each of the three principal US policymakers during the 2008 financial crisis has taken this position at different times. Former US Secretary of the Treasury Tim Geithner suggested that “Financial crises cannot be reliably anticipated or pre-empted.” According to former US Secretary of the Treasury Hank Paulson, “My strong belief is that these crises are unpredictable in terms of cause, timing, or the severity when they hit.” According to Federal Reserve Chairman Ben Bernanke, “This crisis involved a 21st century electronic panic by institutions. It was an old-fashioned run in new clothes.” This line of thought is further supported by early empirical studies showing that, even if most crises are preceded by weak economic fundamentals, they are not especially predictable. If these views are correct, then policymakers should concentrate their efforts on ‘firefighting’, or cleaning up after a crisis. But policing market conditions before crises actually happen would be futile.
An alternative view sees financial crises as predictable enough that policymakers should try to prevent or mitigate them ex ante. This view sees financial crises as the outcomes of overheated credit markets, characterized by rapid expansions of credit accompanied by asset price booms (Minsky 1977 and Kindleberger 1978). Borio and Lowe (2002) show that rapid credit growth and asset price growth predict banking crises in 34 countries between 1970 and 1999, spurring numerous academic policy studies on so-called ‘early warning indicators’. More recently, Schularick and Taylor (2012) and others have shown that credit expansions and narrow credit spreads predict financial fragility.
Even with all this evidence, precise estimates of the probability of a financial crisis following credit and asset price booms remain unavailable. In Greenwood et al. (2020), we draw on newly available crisis chronologies and data to estimate the probability of financial crises as a function of past credit and asset price growth.
Our analysis is facilitated by a chronology of financial crises developed by Baron et al. (2020). They use hand-collected historical data on bank stock returns to improve existing crisis chronologies, which to date had been based primarily on narrative accounts. We combine their list of financial crises with data on the growth of outstanding credit to non-financial businesses and households and with data on the growth of equity and home prices, to estimate the future probability of a financial crisis in a panel of 42 countries between 1950–2016.
Our first finding is that if there is a large credit expansion with an asset price boom, then financial crises are in fact highly predictable. When either non-financial business credit growth is high and stock market valuations have risen sharply, or when household credit growth is high and home prices have risen sharply, the probability of a subsequent crisis is substantially elevated. This is shown in Table 1, where we list the probability of a financial crisis occurring within three years as a function of past credit and asset price growth. The probability of a crisis beginning within the next three years is 45% when equity price growth is in the top tercile of its historical distribution and business credit growth is in the top quintile of its historical distribution. When home price growth is in the top tercile and household credit growth is in the top quintile, the probability of a crisis beginning within three years is 37%.
Table 1 Probability of a financial crisis as a function of past credit and asset price growth
We use these results to motivate a simple indicator variable called the Red-zone, or the ‘R-zone’ for short, that identifies periods of potential credit-market overheating. A country is in the ‘business R-zone’ if non-financial business credit growth over the past three years is in the top quintile of the historical distribution, and stock market returns over the same window are in the top tercile. Similarly, a country is in the ‘household R-zone’ if household credit growth over the past three years is in the top quintile of the historical distribution, and stock market returns over the same window are in the top tercile. Using these R-zone predictors, we show that crises are predictable but slow to develop, suggesting that policymakers have time to act based on early warning signs. For example, the US was in the household R-zone in 2002-2006, a clear harbinger of the crisis that started in 2007.
Credit market overheating often occurs in different countries at the same time. We show that when many countries are in the R-zone, the probability that all countries experience a crisis in the coming years is higher. For example, while Germany was nowhere near the R-zone in 2007, 33% of countries were in the business R-zone and 36% were in the household R-zone at the time. As a result, the predicted probability of experiencing a crisis within 3 years was 37% for Germany in 2007 and, indeed, Germany experienced a crisis in 2008.
Does our finding that the probability of a crisis sometimes rises above 40% warrant the conclusion that crises are predictable? Is 40% enough to warrant pre-emptive action – such as tightening monetary policy? A sceptic might say no. After all, even when starting in an R-zone, which occurs in less than 10% of all country-years, it is far from certain that a crisis will occur. And, as Figure 1 shows, the R-zone does not catch all crises, even if it does flag a large percentage of them. In this regard, two points are in order. First, the economic cost of crises is enormous: the typical crisis leads to permanent reductions in real GDP. Second, we reached our conclusions with just two country-level variables—past credit growth and asset price growth. Just adding the global version of our R-zone indicator would increase predictability even more.
Hence, our conclusion is that financial crises are sufficiently predictable for early action in response to credit market overheating to generate substantial benefits. Our evidence supports the view that the economic system is vulnerable to predictable boom-bust cycles driven by credit expansion and asset price growth. This view suggests that policymakers should pursue prophylactic policy interventions that lean against the wind. Indeed, the post-global financial crisis era has witnessed the advent of several macroprudential tools that have recently been used to counteract credit expansion, including the introduction of time-varying bank capital requirements under Basel III, and the increased use of time-varying maximum loan-to-value standards. Hence, a little more policing, and a little less firefighting, can help foster financial stability.
Figure 1 Financial crises in and out of the R-zone
Baron, M, E Verner and W Xiong (2020), “Bank equity and banking crises”, working paper.
Borio, C and P Lowe (2002), “Asset prices, financial and monetary stability: exploring the nexus”, BIS working paper series, No. 114.
Greenwood, R, S G Hanson, A Shleifer, and J A Sørensen (2020), “Predictable Financial Crises”, Harvard Business School working paper, No 20-130.
Kindleberger, C P (1978), Manias, Panics and Crashes, New York: Basic Books.
Minsky, H (1986), Stabilizing an Unstable Economy, New Haven: Yale University Press.
Schularick, M and A M Taylor (2012), “Credit booms gone bust”, American Economic Review 102: 1029-1061.