Can we optimize an Early Warning System based on earthquakes to improve investments?

In this paper we investigate whether we can optimize an EarlyWarning System based on earthquakes, proposed in Gresnigt et al. (2015), to improve investments. We simulate 1000 asset return series with a GARCH(1,1) model and with a GJR(1,1) model and use the most negative returns in each series to estimate parameters in an Epidemic-Type Aftershock Sequence model. These parameter estimates are used to estimate the probability of a crash in the next five days. We then propose an investment strategy that only invests in the assets for which the estimated crash probability is smaller than some threshold. We explore several ways to determine the optimal threshold and find that it is indeed possible to optimize our Early Warning System. This is done by defining crashes as the 95% quantile of negative returns and choosing the threshold that maximizes the average Hanssen-Kuiper Skill Score over all simulated returns. In particular, this strategy yields a Sharpe ratio that is approximately twice as high as the Sharpe ratio of a strategy that remains invested in all assets.