Zombies are the least of your worries during a zombie apocalypse.

After you come to terms with the initial shock of waking up to your zombifying girlfriend, the real danger is other humans (cf. many movies and TV shows).

Most large stock crashes are basically the financial equivalent of zombie apocalypses.

Once the initial shock is over - start worrying about humans.

Your dad hears about the hoo-ha and sells a day later. The penny drops with granddads everywhere about a week later, the initial shock turns into ripples of shocks.

Academic finance is predicated on efficient flows of information, but when you have delays in reactions, markets become far less efficient and shudder to and fro. This is why simple mean reversion works.

Recently the League of Justice strategy made a miraculous recovery, and it looks like another strategy is making a comeback from beyond the grave.

Steady Vol inversely weights holdings in an index by the inverse of the current volatility, it's a timing strategy, i.e. during highly volatile periods it jumps into cash and vice versa.

It did pretty badly when the volatility in the weighting calculation is calculated using 10 daily returns (Big 'O' / 'worst case' Sharpe of 0.35).

But how about calculating realised volatility with a week's, a month's and a quarter's worth of returns?

What do we find?

As expected, volatilities that are calculated with fewer daily returns have a larger range of Sharpes. They tend to bump around wildly as we change the inputs a little.

The Steady Vol strategy calculated using just 5 daily returns to calculate volatility covers a lot of ground - it's been tested more robustly than the others.

The 'high' run in this example rebalances on the 19th of every month and the 'low' around the 11th.

So while we have tested using a history of 25 years, we have generated 21 backtests instead of one, with a wide variety of scenarios within that range. The 0.26 Big 'O' Sharpe is very low, but on the upside it's an extremely robust lower bound(!).

Using 21 days of returns in our return calculation gives us a much less of a spread between highs and lows and therefore less varied scenarios, but it does give us a pretty decent lower bound of 0.55 (versus 0.4 for buy and hold S&P 500).

Using just 5 returns to calculate volatility serves to prove a point. You need to balance meaningfulness and robustness. Nevertheless it gives you an idea of the benefits of using less data when calculating your signals and reducing dependence in your strategy returns.

Steady Vol Just Won't Die

You have to be careful about overfitting parameters, but the robustness check and the fact that we rebalance every month feels like a good fit.

Code for the Lazy Backtesting IDE is here.