Choosing Forecast Targets for Market-Taking High-Frequency Trading
Summary
The document explains that a useful forecast target depends on the market-taking task. For a large order, a short-horizon mid-price forecast can help time execution. For a simple single-exchange strategy, the relevant question is whether prices will cross the bid or ask before the intended exit, while accounting for entry slippage. Cross-exchange arbitrage may focus on top-of-book prices across venues; structural arbitrage across several exchanges makes slippage a central concern.
The answer is conceptual rather than empirical: it gives no model, dataset, or performance results. It also does not settle whether weighted mid-price or recent trade-price changes are generally predictive. Instead, it frames target selection around the intended trade and execution costs, with a brief acknowledgment that the explanation is simplified. Practical forecasts would need to reflect order size, venue, horizon, and realized execution conditions.
Key ideas
- A forecast target should match the trading task and intended order.
- Short-term mid-price forecasts can help time execution of large orders.
- Single-venue market taking requires comparing expected exit prices with entry slippage.
- Cross-exchange arbitrage may use top-of-book prices, while multi-venue structural arbitrage must account for slippage.
- The answer is simplified and offers no empirical validation.
Tags
Full text
# Target variables in high frequency trading
# Target variables in high frequency trading
Given that we are a market taker (removing liquidity from the limit order book through market orders), what should we be trying to forecast?
It seems like the most pertinent thing for us to forecast is the execution prices of our buy or sell orders of a certain size so that we can determine if our current entry and future exit would be profitable, e.g. the mid price and spread for sufficiently small orders, but then what are constructions such as the weighted mid price $(\frac{p_\text{bid}q_\text{offer} + p_\text{offer}q_\text{bid}}{q_\text{offer} + q_\text{bid}})$ used for? Are they meant to be predictive of future mid prices/potentially spreads?
I saw that this answer describes how features such as the last traded price (or perhaps the log returns of the last traded price) can be used to predict log returns of the mid price which is also described by that answer as being relevant for market taking, but wanted to confirm if this is what we're usually trying to predict in a market taking context.
## Answer by Alex D (score 3)
https://quant.stackexchange.com/a/76554
What you need depend on what you are trying to do (note that the rest of the explanation is very simplified).
If your target is to execute a big order, you can use a short term mid price prediction to time your actions (why buy at $P_{T}$ when you can buy at $P_{T+\Delta t}$, where $P_{T} > P_{T+\Delta t}$).
If you are doing a naive HFT on a single exchange - you are interested in a bid/ask price crossing between $T$ and $T+\Delta t$. Here you are interested in predicted exit price with respect to a slippage on enter.
If you are trying to arbitrage between exchanges - you may try to predict ToB (top of the book) prices on both of them and try to cross them.
If you are doing structural arbitrage on multiple exchanges - you are very interested in slippage.Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.