Do equity markets respond predictably to the probability and fact of war? In their May 2011 paper entitled “The War Puzzle: Contradictory Effects of International Conflicts on Stock Markets”, Amelie Brune, Thorsten Hens, Marc Rieger and Mei Wang relate U.S. stock market returns to the estimated likelihood of war involving the U.S. as evidenced by analysis of news (key word counts in the New York Times, with one of the key words being “war”). They consider the six wars most costly to the U.S. since World War II (World War II, Korean War, Vietnam War, Gulf War, Iraq War and Afghanistan War). They distinguish between wars that have obvious preludes and wars that surprise (with the Korean War the paradigm for the latter). Based on availability of multiple measures of probability of war, they use the onset of the Iraq War during late 2002 through early 2003 to benchmark the stocks-war relationship. Using New York Times news reports during the selected wars and contemporaneous levels of the S&P 500 Index and the Dow Jones Industrial Average, they find that:
- When there is a pre-war phase, an increase in the likelihood of war tends to depress the equity market, but the ultimate outbreak of a war tends to elevate the market.
- When war surprises, the outbreak tends to depress the equity market.
- A possible explanation is variance aversion in a mean-variance context, the interplay of return expectations and uncertainty in those expectations.
In summary, evidence suggests that investors should be bullish (bearish) at the onset of a war that has an identifiable prelude (that surprises).
Cautions regarding findings include:
- The sample of wars is small, inhibiting statistical inference and representing few investment opportunities. Segmenting the sample into “prelude” and “surprise” subsamples exacerbates this sample size constraint.
- There is arguably look-ahead bias in the selection of war prelude and outbreak test intervals. For example, an investor operating in real time would have had difficulty determining these intervals for the Vietnam War. In other words, identifying these intervals out of sample may be problematic.