Signal-Based HFT

A common thread between the two archetypal quant strategies discussed so far is the presence of a fair price. In fact, arbitrage can be reframed as a strategy on the product in the less liquid venue, using the mid-price of the more liquid venue as the fair price. In this way, the arbitrage strategy is equivalent to crossing the book whenever our fair price crosses the market.

Under this framework, a problem becomes obvious: what if the fair price moves in a direction such that we lose money on a trade before we have had a chance to complete the risk-offsetting trade? For example, in an arbitrage strategy, I buy a security on a less liquid venue, only for the price to drop on the more liquid venue before I can sell it there. Or, in a market-making strategy, I buy, but the price drops before I can sell at my current offer price. The problem is that the fair price is not predictive.

Currently, we have just been using a mid-price as the fair price. Over time, quant firms have built up intuitions about what the order books look like when the mid-price is about to go up or down. These short-term predictors can, in a modern ML-enabled world, be considered features in a more general model that determines the movement of the traded security over a short horizon. This is the current state of the art in high-frequency trading, with firms employing hordes of quants to produce more accurate pricing, often combining aggressive and passive trading strategies into one.