decline in a national stock market index. We identify 1,032 events for which a market declined by more
than 50% over a 12-month period. Conditioning on these events, and controlling for a range of other
factors, we find that markets tend to rebound in the year following the crash. We refer to this pattern of
crash-and-rebound as a ‘negative bubble.’Interestingly, the pattern only holds for large crashes −
declines of lesser magnitude exhibit persistence, not reversal. This non-linearity presents a challenge to
standard econometric forecasting techniques and suggests that something more complex than mean
reversion is at work.
We consider some proposed macroeconomic, institutional, and sample selection explanations for
this pattern. Macroeconomic explanations include changing volatility, inflation risk, and disaster risk
as potential drivers of temporary asset price declines. Institutional explanations include financial sector
fragility and constraints on arbitrage. Selection bias theories posit that ex post data collection omits
markets that failed to rebound. We find negative bubbles are not well explained by macroeconomic or
institutional factors. Selection bias also does not seem to drive the results. Alternatively, there are
behavioral theories that are consistent with the findings. They include panics or other temporary shocks
to sentiment, and bubbles −temporary, irrational over-pricing leading to a crash. While promising,
they are difficult to test.
There is a long literature of forecasting long-horizon stock market returns, much of it modeling a linear
relationship between a set of macroeconomic state variables and future returns. Although the
limitations of this approach have been studied and discussed, the consensus is that some combination of
past returns, financial ratios, and discount rate proxies have predictive power. This large body of work
seeks to identify some model or consistent rules that govern market behavior under a complete range of
conditions. Our approach in this paper is different in that we focus narrowly on extreme events in order
to document a comprehensive set of outcomes. We then explore varying conditions that may have
affected outcomes, and compare these to proposed theories. Documenting historical outcomes may not
provide a complete guide to predicting future consequences of a stock market crash, but it is better than
relying on personal experience, or selective anecdotes about prior crashes. Thus, our modest goal is to
report the frequency of market crashes and the distribution of subsequent returns.
Our second goal is to evaluate some of the theories proposed to explain the dynamics of markets
following crashes. Muir (2017) documents a V-shape in asset prices specifically associated with
financial crises over 140 years of history and 14 markets. Notably, this pattern is not associated with
more fundamental factors such as consumption shocks or other major macroeconomic events. Rather,
asset price dips may reflect distress caused by over-leveraged financial institutions (cf. Brunnermeier
& Sannikov, 2014). Financial constraints have been shown to cause fire sales in asset prices, which are
then followed by a rebound (cf. Brunnermeier & Pedersen, 2008; Geanakoplos, 2010; Gromb &
Vayanos, 2009; and Richardson, Saffi, & Sigurdsson, 2017). Coval and Stafford (2007) document this
pattern for stocks held by mutual funds experiencing large outflows. In short, both asset supply and
demand are affected by institutional distress. On the supply side, banks and arbitrageurs may be forced
to de-risk by selling assets. On the demand side, distressed banks may restrict credit to arbitrageurs in
periods when assets are priced cheap.
We test these hypotheses in two ways. First, we test whether crashes coincident with financial
crises are more or less likely to rebound.
We find little evidence that financial crises matter when we
We rely on the assumption that we have a comprehensive set of financial crises for the countries we study.
GOETZMANN AND KIM