Short-Term Reversal (Cross-Sectional, Weekly)

Thesis (edge)

Over horizons of about a week, stocks that have just done badly tend to do slightly better than average next, and stocks that have just done well tend to do slightly worse. This is the opposite of momentum, which works over months, and it is one of the oldest documented patterns in equities.

The reason is not that the market is stupid. It is that a lot of short-term price movement is not information, it is pressure. A large fund liquidates a position. An index change forces selling. A risk desk cuts exposure. Whoever must trade in a hurry pushes the price away from fair value, and someone has to take the other side. The person who takes the other side is providing liquidity, and gets compensated when the price drifts back.

So the honest framing is this: short-term reversal is not a prediction about companies. It is a fee you collect for absorbing other people's urgency. That framing matters, because it tells you exactly when the strategy will hurt you.

Where it works (regimes)

• Works well: in normal, liquid markets with steady two-way flow, where the pressure that pushed a stock down was mechanical and reverses quickly.
• Works well: in more liquid names, where you can actually trade the signal without paying away the entire edge.
• Fails badly: in a genuine crisis. When everyone wants to sell at once, being the buyer of the worst performers is not a clever contrarian bet, it is standing in front of the flow. The strategy has its largest drawdowns precisely when volatility spikes.
• Fails: when the move was information. A stock that fell twenty percent because it lost its biggest customer is not going to bounce because the last five days were negative.
• Decays: the raw version is extremely well known, and the gross returns have shrunk over decades as electronic market makers have made liquidity provision cheaper and more competitive. What was once a strategy is now, in liquid names, largely a market maker's business.

Signals

The core signal is embarrassingly simple.

• Rank stocks by last week's return. Buy the bottom decile, sell the top decile. That is it.
• Filter by liquidity. The raw signal is strongest in illiquid names, and that is a trap, because illiquid names are where you cannot trade it. Apply a hard liquidity floor and accept the weaker signal in tradable names.
• Exclude event-driven moves. If the stock moved because of earnings, guidance, a deal or a regulatory decision, that is information and it is not going to revert. Excluding earnings windows materially improves the quality of the signal.
• Consider skipping the most recent day. If a stock is being actively liquidated today, buying it today means standing in front of a seller who is not finished. A short skip can improve entry.

Portfolio construction

• Market neutral by construction: equal capital long and short, so you are not making a directional bet.
• Better still, factor neutral: neutralise sector, size and beta. Without this, the strategy can accidentally become a bet on whatever sector had a bad week, which is not the effect you are trying to harvest.
• Many small positions: this signal is weak per name and only works as an average over a large number of names. A concentrated version of this strategy is not a smaller version, it is a different and worse strategy.
• Volatility scaling: give less weight to names whose price swings wildly, or the book will be dominated by a handful of jumpy small caps.
• Rebalance weekly, and expect the turnover to be enormous. Most of the book turns over every cycle.

Risk model

• Cost risk is the main risk. Gross returns can look excellent. Net returns after realistic spreads, impact and borrow are often close to zero in liquid names. This is not a footnote, it is the central fact of the strategy.
• Adverse selection: the reason a stock is cheap may be that someone knows something. When you systematically buy from informed sellers, you lose. This is the permanent cost of being a liquidity provider.
• Crisis risk: the payoff profile resembles selling insurance. Small, steady gains most of the time, with occasional sharp losses when markets dislocate. Size accordingly, and do not mistake the calm periods for low risk.
• Crowding: many quant funds run some version of this. When they deleverage together, the reversal book gets hit at the same moment, which is exactly what happened in the quant quake of August 2007.
• Borrow: the short book is drawn from the biggest recent winners, which can include names with expensive or unavailable borrow.

Costs & implementation

This section is not optional here. It is the strategy.

• Turnover is close to complete every week. You pay the spread on essentially the whole book, twice, every cycle. At fifty cycles a year, even a few basis points of cost per trade compounds into a very large number.
• Impact grows with size. The strategy's capacity is limited, and it is limited precisely in the small and mid caps where the signal is strongest.
• Borrow is a real cost and the short side determines a lot of the net result.
• Execution matters more than signal refinement. Passive, patient execution using limit orders is not a nice-to-have. Since you are being paid to provide liquidity, taking liquidity aggressively to enter the trade actively destroys the source of the edge.
• Report net, always. Any presentation of this strategy that shows gross returns should be treated with suspicion.

Failure modes

• Backtesting gross and celebrating. The single most common error. The gross equity curve is beautiful and irrelevant.
• Letting microcaps into the universe. They inflate the backtest and cannot be traded.
• Ignoring earnings and news. Buying stocks that just collapsed on real news is not reversal, it is value destruction on a schedule.
• Sizing it like a normal strategy. The return distribution is negatively skewed. Sizing to average volatility means being far too large during the events that actually matter.
• Aggressive execution. Crossing the spread to buy the falling stock immediately turns the liquidity premium you were trying to earn into a cost you pay.
• Using stale price data. Illiquid names with stale closing prices generate fake reversals that disappear the moment you try to trade them.

Our Notes & Suggestions

Build the cost model before you build the signal. This is the rare strategy where the ranking rule takes an afternoon and the cost and execution work takes months. If your infrastructure cannot execute patiently and cheaply, the correct conclusion is not to run a worse version of the strategy, it is not to run it at all.

The most useful upgrade is to stop reverting raw returns and start reverting residual returns, which strips out the market, sector and factor moves and isolates the part that is genuinely idiosyncratic pressure. That is a distinct strategy and is covered separately.

Finally, respect the shape of the payoff. You are being paid a small, regular premium for taking the other side of urgency, and every so often the urgency is right and you are wrong, in large size, all at once. Any risk framework that treats this as a steady low-volatility return stream will eventually be surprised.