Stock Market Back testing through Simulation

Could we predict how the stock market will change in the future? Could we foresee arising risks and opportunities by simulation? How precise would that be? These are questions of interest for any investor and Tony Cooper, from Double Digit Numerics, makes an in-depth analysis on this phenomenon in the paper called “Simulation as a Stock Market Backtesting Tool”, available at SSRN: SSRN or at SSRN.

Simulation in the hi-tech era

The exponentially increasing power of computers offers a new form of experimentation which lets us perform entire experiments on computer only, called “in silico” experiments. This goes as far as replacing “in vivo” experiments, for example testing drugs on computer models on mice instead of testing them on real mice. Therefore, recently, a new class of antibiotics – called oxadiazoles, has been discovered on computer, through “in silico” experiments.

So how could we use all that to predict stock dynamics?

We currently test investment strategies on past market data, hoping to get some clues for building our performance strategy on future data. But, often we do not possess enough past data to get relevant statistical conclusions on future market performances and, even if we would, it is unlikely for the future market to resemble to the past one.

The paper analyzes how we can use simulation to generate future market data to test different hypothesis and strategies. The proposed simulator models markets in different ways, including algorithms for generating noise and random walks of various kinds, like trending and mean-reverting walks. It permits time variation to all included parameters, such as switching between trending and mean reverting phases, whole stock portfolios and Exchange Trade Funds incorporation, and regime changes generation, like switches from volatile to calm markets.

Optimistic perspectives

The simulator’s improving possibilities are infinite, as it is an open source app. Different examples and simulations are performed and described, while seeking to match the “in silico” environment to the “in vivo” one. Finally, it is proven that even if the simulated result may not be highly accurate, they are definitely indicative. With new algorithms coming out every week and by leaving open others’ possibility to bring their own contribution to improving the simulator, we can expect even more spectacular results in the stock market back testing field, for the very near future.