few minutes after 1 p.m. Pacific time on April 23, 2013, the U.S. equity markets fell sharply and within a few minutes recovered completely (see Figure 1). One would miss this event completely if one were looking at closing market prices. News articles later attributed the sharp fall to a rumor created by a hack on a news service’s Twitter feed. For those managing the tail risks of investment portfolios, this rerun of the “flash” crash provided a real-time laboratory for identifying where market stresses might lie. High frequency events can also be a valuable source for statistics on rare events. Whereas rare events that occur over lower frequency, longer horizons are much harder to find (and hence much harder to derive statistics from), intraday events create a larger, more accessible data set that can be used to supplement data on tail events.
Analyzing the reactions of different markets to this intraday tail event can provide valuable information for investors looking for effective tail risk hedges for their portfolios. Figure 2 illustrates the simple, linear correlations of various markets to the S&P 500 index on April 23. There are a few things that stand out. First, the event correlations during the stress are all higher than the correlations excluding the stress event. This makes sense, since a macro news item can affect all markets simultaneously. Second, most assets show risk-on or risk-off behavior: That is, their correlations are either (1) positive to the equity market’s flash move, if they are risk-on, as in the case of the Australian dollar, where a currency carry trade is favored; or (2) negative to the equity market, if they are risk-off, as in the case of the Treasury bond market, which has historically been seen as a safe haven during crises and negatively correlated to equities. There are also some assets that switch from being uncorrelated to correlated (e.g., crude oil futures). The case of the yen is especially impressive as it strengthened substantially during the event (see Figure 3), likely because it is on the other side of carry trades, almost matching the equity market’s sharp fall. Third, we find that the full range of the day’s market movements in an otherwise low volatility day was dominated by the sharp movements around the event. What this indicates to us is that while market volatility might be low by historical standards, the magnitude of correlated fat tail events is not a direct consequence of low volatility.
A similar spike of correlations also happened on the morning of Friday May third, when the U.S. nonfarm payroll data came in early before the New York Stock Exchange opened (Figures 4 and 5). The data surprised the financial markets on the upside and markets reacted fast in the ten minutes afterwards.
Why does this dynamic arise? We would speculate that markets that show high and positive correlation to equities are also the ones that are held by levered investors, who are quick to abandon these markets in adverse shocks. If these holdings are funded by low interest rates – the yen carry trade (i.e., an implicit short position in the yen), for instance – then much of the sign and magnitude change of the correlation makes sense.
Given this anecdotal evidence from the response of markets to this mini laboratory experiment, investors can do three things. First, stay away from significant exposures to markets that show a large and increasing response to risk-off behavior, unless adequately compensated for it; second, look to indirect ways of hedging tail risks from asset classes that show low correlation to equities and where options are attractively priced; and third, while some diversification is critical, don’t put too much faith in diversification working too well in tail events.