Advances in Financial Machine Learning

CHAPTER 13 Backtesting on Synthetic Data

13.1 Motivation

In this chapter we will study an alternative backtesting method, which uses history to generate a synthetic dataset with statistical characteristics estimated from the observed data. This will allow us to backtest a strategy on a large number of unseen, synthetic testing sets, hence reducing the likelihood that the strategy has been fit to a particular set of datapoints.¹ This is a very extensive subject, and in order to reach some depth we will focus on the backtesting of trading rules.

13.2 Trading Rules

Investment strategies can be defined as algorithms that postulate the existence of a market inefficiency. Some strategies rely on econometric models to predict prices, using macroeconomic variables such as GDP or inflation; other strategies use fundamental and accounting information to price securities, or search for arbitrage-like opportunities in the pricing of derivatives products, etc. For instance, suppose that financial intermediaries tend to sell off-the-run bonds two days before U.S. Treasury auctions, in order to raise the cash needed for buying the new “paper.” One could monetize on that knowledge by selling off-the-run bonds three days before auctions. But how? Each investment strategy requires an implementation tactic, often referred to as “trading rules.”

There are dozens of hedge fund styles, each running dozens of unique investment strategies. While strategies can be very heterogeneous in nature, tactics ...