Common algorithmic trading strategies

Strategy families

There are countless individual strategies, but most belong to a handful of families. Each one earns its return by taking a particular kind of risk or providing a particular service to the market. Understanding that trade-off is more useful than memorising names.

Trend following and momentum

Trend-following strategies buy assets whose prices have been rising and sell those that have been falling, on the view that persistent moves tend to continue for a while. They are simple to state and have a long history across futures, currencies and, more recently, digital assets. They tend to do well in long, directional markets and poorly in choppy, range-bound ones, where they are repeatedly caught by false breakouts.

Mean reversion

Mean-reversion strategies take the opposite view over shorter horizons. When a price moves unusually far from a reference level, such as a moving average, the strategy bets on a partial return. These strategies often win frequently in small amounts. The danger is a genuine regime change, where the price does not come back and the position keeps losing.

Statistical arbitrage

Statistical arbitrage looks for stable relationships between related instruments. The classic form is pairs trading: two assets that usually move together drift apart, so the strategy buys the cheaper one and sells the richer one, expecting the gap to close. Larger versions trade hundreds of instruments at once. The main risk is that a relationship which held historically breaks down, often at exactly the moment it is most heavily traded.

Market making

A market maker quotes both a bid and an offer and aims to earn the difference between them as other participants trade. It provides liquidity that makes markets cheaper for everyone else. Its risks are inventory, since it can be left holding a position when prices move, and adverse selection, since the traders who hit its quotes may know something it does not. Successful market making depends on fast quote updates and careful inventory control.

Cross-venue arbitrage and liquidity rebalancing

When the same asset trades on several venues, prices can differ briefly. A cross-venue strategy buys where the asset is cheaper and sells where it is dearer. In practice, moving assets between venues takes time and costs money, so these strategies hold inventory on each venue in advance and rebalance it periodically. Returns per trade are small and competitive, and the work lies in fees, transfer logistics and keeping inventory within sensible bands.

Basis and funding trades

Derivatives are linked to their underlying asset but do not always trade at the same price. In digital-asset markets, perpetual futures use a periodic funding payment between long and short holders to keep their price close to spot. A trader who holds the spot asset and an equal short perpetual position is broadly neutral to price, and receives funding when it is positive. The risks are venue counterparty exposure and the need to hold enough margin so the short leg is never liquidated in a sharp rally.

Execution algorithms

Execution algorithms do not seek a view on price; they aim to complete a given order efficiently. Common designs include:

TWAP

Time-weighted average price. Spreads the order evenly across a time window.

VWAP

Volume-weighted average price. Trades in proportion to expected market volume, so more is done when the market is busiest.

POV

Percentage of volume. Keeps participation at a fixed share of traded volume.

Implementation shortfall

Balances the cost of trading quickly (market impact) against the risk of the price moving while waiting.

Even these simple tools need limits. The joint CFTC and SEC report on the 6 May 2010 Flash Crash described a sell program of 75,000 E-mini contracts, worth about $4.1 billion, executed by an algorithm set to target 9% of trading volume “without regard to price or time”. As liquidity thinned, the algorithm kept selling into a falling market.

Machine-learning approaches

Machine learning can find non-linear patterns across many inputs, and it is widely used for forecasting short-term price moves, classifying market regimes and improving execution. It also magnifies the classic research problems. Financial data is noisy and non-stationary, and a flexible model can fit noise convincingly. Careful validation, discussed in the backtesting article, matters even more here.

Summary

Trend following

Needs persistent moves. Main risk: choppy, directionless markets.

Mean reversion

Needs temporary dislocations. Main risk: a lasting regime change.

Statistical arbitrage

Needs stable relationships. Main risk: correlations breaking down.

Market making

Needs two-way flow. Main risk: inventory and adverse selection.

Cross-venue

Needs price differences larger than costs. Main risk: fees, transfer delays, venue risk.

Basis and funding

Needs a persistent premium. Main risk: margin calls and counterparty exposure.