Tactical Investment Algorithms
11 Pages Posted: 30 Sep 2019 Last revised: 1 Oct 2019
Date Written: September 26, 2019
There are three fundamental ways of testing the validity of an investment algorithm against historical evidence: a) the walk-forward method; b) the resampling method; and c) the Monte Carlo method. By far the most common approach followed among academics and practitioners is the walk-forward method. Implicit in that choice is the assumption that a given investment algorithm should be deployed throughout all market regimes. We denote such assumption the “all-weather” hypothesis, and the algorithms based on that hypothesis “strategic investment algorithms” (or “investment strategies”).
The all-weather hypothesis is not necessarily true, as demonstrated by the fact that many investment strategies have floundered in a zero-rate environment. This motivates the problem of identifying investment algorithms that are optimal for specific market regimes, denoted “tactical investment algorithms.” This paper argues that backtesting against synthetic datasets should be the preferred approach for developing tactical investment algorithms. A new organizational structure for asset managers is proposed, as a tactical algorithmic factory, consistent with the Monte Carlo backtesting paradigm.
Keywords: Backtest overfitting, selection bias, multiple testing, quantitative investments, machine learning, all-weather hypothesis, strategic investment algorithm, tactical investment algorithm.
JEL Classification: G0, G1, G2, G15, G24, E44
Suggested Citation: Suggested Citation