Around mid-February, equity markets reached new all-time highs, volatility was reasonably low and credit markets functioned properly. Ever since then, due to the COVID-19 outbreak, global equity markets dropped by as much as 30%, volatility indices spiked and default risk increased. Many quantitative funds (QFs), in particular, have suffered substantial losses during the market turmoil. This even holds for funds that offer market-neutral quantitative strategies.
Credit Suisse estimates that QFs have almost halved their positions at the start of the previous month and that the average QF lost 14% percent. To illustrate, Renaissance, a well-known QF, has seen losses deepen in March, with its flagship fund (Renaissance Institutional Equities fund) losing 18% in March, and year-to-date 24%.
They say that one should “never waste a crisis”, which to me, means that valuable lessons should be drawn from the current pandemic. As a quant, I would like to share three lessons that I learned during this crisis: Forget forecasting, nowcasting is the future (1), value the strength of economic theory (2), and there exists no Holy Grail (3). Allow me to elaborate further, point by point.
1: Forget forecasting, nowcasting is the future:
Traditional forecasting models utilize data that is structured to make long-range predictions, beyond quarters and years. These models rely on statistical relationships between past observations and future realizations. However, such relationships tend to be time-varying or may even disappear. Quant strategies may be based on predicting asset prices/returns based on time-series dynamics (for example, in statistical arbitrage and CTAs) or based on cross-sectional data (for example, asset pricing and factor investing). To illustrate the problem here, factor investing strategies may determine the value of an asset as a function of their exposure towards asset pricing factors, which happens to be reported infrequently. These factor exposures are not updated frequently enough to keep up with the current market conditions.
Nowcasting techniques, in contrast to forecasting, use datasets that are unstructured to make direct measurements (the target variable is observed) or short-term predictions (the target variable is not directly observed). One advantage is that direct measurements always hold true since they do not rely on a statistical relationship. Another advantage is that short-range predictions appear to be more reliable and timely than long-range predictions. Practical applications of nowcasting in finance are plausibly endless, including predicting inflation (without using econometric models but rather web-scraping), liquidity conditions (by using data from market makers in real-time), and industrial production.
Many people argue that the current sell-off due to the pandemic was a Black Swan event. I tend to disagree here. Those who implemented nowcasting techniques may have observed plenty of early warning signs that the virus was disrupting supply chains in China. The sell-off may be a Black Swan event to market forecasters, but to nowcasters, this is a simple White Swan event. A wise lesson that I have drawn here is that nowcasting techniques should be included in the arsenal of the quant, preferably complementing traditional techniques.
2: Value the strength of economic theory:
One of the favorite tools of the Quant is backtesting, which is a useful tool, when conducted properly. In a previous blog, I already gave a rant about the dangers of backtesting. When one would run thousands and thousands of historical backtests – believe me or not, this is not a lot – you can find a lucky factor and identify a promising investment strategy, which can be used for publication or to launch a new quantitative fund. It is trivial to construct a historical backtest with a nice Sharpe ratio by trying out a thousand alternative model specifications. Consequently, this may explain why many published discoveries in finance are false, and why many quantitative funds do not perform as well as we would expect based on backtests. The latter is something that we observe right now during the COVID-19 crisis.
Backtesting should be merely used for validation. However, backtesting has been used by many to build trading rules and form hypotheses rather than attempting to refute a hypothesis. Misuse of backtesting can be used to circular arguments. For example, a researcher backtests thousands of trading strategies (i.e. the factor zoo). The strategy with the highest Sharpe ratio is proposed as a hypothesis (i.e. long past winners, short past losers). Then the researcher publishes this hypothesis and uses the same backtest (the one that was used in first instance to form a hypothesis) as evidence in favour of the hypothesis. Researchers should focus on developing theories, independent from backtesting, that explains a phenomenon by an exact cause-effect mechanism. Subsequently, this validity of this mechanism should be tested through both backtesting and collecting evidence against the implications of these theories.
3: There exists no Holy Grail:
Academics and practitioners typically tend to develop strategies and solutions with a “one-size-fits-all” mindset, i.e. something that performs well across all market regimes (a “Holy Grail”). Most strategies are backtested over decades (some studies even use data going back multiple centuries) to show that a strategy worked in the past across different macroeconomic settings. Yet these long histories may not be representative of the current environment (for example low interest rates) or for what yet has to come.
Besides, the likelihood that a real strategy that always works is rather slim. Even if they do exist, they compromise a small subset of the population of strategies compared to strategies that work across one or a few regimes. Looking for such a strategy would be similar to finding a needle in a haystack.
Quants should instead focus their efforts and time to develop investment strategies that perform optimally under a specific or few market regimes. As each market regime has a unique data generating process, we can nowcast the probability of each regime at each point in time. Using these probabilities, one could construct portfolios of those optimal strategies. The advantage here would be that quantitative funds can more timely rebalance their portfolios as market conditions adapt. Regime-specific strategies are common among market markets, whereas quantitative funds tend to be lagging on this respect.