Reversal is a test, not a punchline
Suppose a rule buys when a fast moving average crosses above a slow one and sells when it crosses below. Reversing the events sounds simple: buy on the bearish crossover and sell on the bullish crossover. If the original rule made money, should the reversed one lose the same amount?
Usually not. A long-only backtest moves between stock and cash. Reversing its events changes which intervals are invested, but it does not create a short position. Costs occur on different dates, the first actionable event can change, and an open final position may remain unpaired. Compounding also means two complementary return paths are not exact negatives.
That does not make reversal useless. It makes it a controlled challenge to the hypothesis. If the story says upward momentum should be rewarded, a rule that deliberately owns the opposite state provides a valuable comparison. The comparison can expose vague reasoning, timing errors, regime dependence, or a result that owes more to general market drift than to the intended signal.
Four different things people call a reversed strategy
Swap the entry and exit events
This is the cleanest long-only interpretation. Every original sell event becomes a candidate buy, and every original buy becomes a candidate sell. The reversed strategy still holds either the asset or cash. It owns the intervals the original rule generally avoided.
Invert the position
For a binary long-or-cash rule, inversion means reversed_position[t] = 1 − original_position[t]. This creates complementary exposure after the original state is known. It may differ from simply swapping events at the start and end of a sample because the event-driven state machine has to decide its initial position.
Turn long signals into short signals
A short strategy uses a negative position and gains when the asset falls. That requires borrowing assumptions, short-sale availability, financing, dividends owed, margin, and potentially asymmetric loss constraints. The SEC’s introduction to short sales explains the borrowing, financing, dividend obligations, and loss exposure. It is not equivalent to staying in cash during the original long periods. TickRun’s strategy backtests are long-only, so a swapped TickRun-style rule should not be described as a short strategy.
Negate the recorded returns
Multiplying every original strategy return by −1 is a mathematical transformation, not a faithfully executable reversal. It also mishandles transaction costs: costs should remain costs rather than become gains. Use it only as a diagnostic calculation, never as the reported performance of a tradable rule.
A small example shows the difference
Imagine five close-to-close asset returns: +2%, −1%, +3%, −2%, and +1%. The original rule is exposed for the first three intervals and in cash for the last two. Before costs, its return path uses +2%, −1%, and +3%. A complementary long-or-cash rule uses 0%, 0%, 0%, −2%, and +1% over those same intervals.
Compounding gives the original 1.02 × 0.99 × 1.03 − 1 = +4.0094% and the complement 0.98 × 1.01 − 1 = −1.02%. These are illustrative returns, not measured TickRun results. The reversed result is not the negative of the original. It captures a different subset of asset returns. Both versions may gain if the positive intervals happen to fall in both subsets, and both may lose after their separate trading costs. If cash earns interest, even the flat intervals have a return that must be specified consistently.
The first row needs special care. An indicator may require a warm-up period before either event exists. If the original state starts flat, an exact position complement would start invested, but a swapped-event implementation might remain flat until it encounters the first former sell. Those are two legitimate experiments with different definitions.
How this relates to TickRun’s execution model
TickRun turns raw buy and sell events into a persistent binary position. A buy is accepted only while flat. A sell is accepted only while invested, and a configured minimum holding period suppresses exit events that arrive too early. A suppressed sell is not queued for automatic execution when the holding period ends; a later eligible sell event is needed. The completed position on session t applies to the following close-to-close return, while a position change is charged its transaction cost on the event row.
Because of that state machine, swapping two raw event arrays can change more than their labels. A former sell that occurs while the reversed strategy is already invested is ignored. A former buy may be unable to close the reversed position if it arrives before the minimum hold has passed; that suppressed event is not automatically retried later. Signal delay applies to the new events in exactly the same way as it did to the originals.
TickRun does not provide one global “reverse every strategy” control. Several experimental strategies do have an explicit direction choice—holding during a high condition versus a low condition—and their implementation reverses the relevant entry and exit logic. For a general reversal study, record the modified rule as a separate hypothesis rather than implying that an existing result was flipped inside the interface.
Important distinction: a reversed long-only rule selects the other invested intervals. It does not automatically short the asset, earn the negative of the original return, or remove market-wide upward drift.
What a reversal test can teach you
Whether direction is doing real work
If both directions behave almost identically, the indicator’s directional interpretation may be less important than expected. The result might be driven by time in the market, one unusual episode, or a broad positive drift in the tested stock.
Whether the rule is really a regime selector
An original rule and its long-only complement divide history into two sets of intervals. Comparing them can reveal that one state has higher average returns, lower drawdown, or different volatility. That is more precise than saying one “predicts up” and the other “predicts down.”
Whether execution asymmetry matters
Entry and exit conditions often have different confirmation, buffers, or thresholds. Reversing them may create radically different trade lengths and turnover. If the reversed rule trades far more often, a performance gap may partly reflect costs rather than directional information.
Whether your explanation is flexible after the fact
If the original wins, it may be called momentum. If the reversed version wins, it can be tempting to rename the same test mean reversion and declare success anyway. A prewritten hypothesis prevents that. State which direction is expected to work and what outcome would count against the idea before viewing both results.
Compare the two versions fairly
- Freeze the source signal. Calculate one indicator series from the same adjusted OHLCV observations.
- Define reversal precisely. Choose swapped events, complementary position, or an actual short rule; do not mix them.
- Use the same dates and warm-up. A comparison becomes misleading if one version begins after more information is available.
- Apply the same delay, holding rule, and costs. Charge each version for its own position changes.
- Report exposure and turnover. A return difference is easier to interpret when one strategy was invested 80% of the time and the other only 20%.
- Inspect complete and open trades. TickRun’s completed-trade count excludes an unclosed position even though that position affects portfolio equity.
- Repeat outside the discovery sample. Reversal is another tested variation and therefore adds to the multiple-testing burden.
Compare total return, drawdown, Sharpe ratio, completed trades, transaction costs, and the full equity paths. Avoid declaring the original superior merely because it has the larger final value. It may also be exposed longer, suffer deeper losses, or depend on one market regime.
Three possible outcomes
The original is clearly stronger. This supports the intended direction within the tested sample, but it does not prove prediction. Check nearby parameters, other stocks, and later data before treating the difference as durable.
The reversed rule is stronger. The original directional story is challenged. The right response is to investigate why, not quietly adopt the reversal as though it had been proposed in advance. It is a new exploratory strategy that needs independent validation.
Both are weak. That can be the most useful answer. The indicator may not divide the sample into economically meaningful states after costs. Recording that result prevents more parameter searching from disguising a failed premise.
Common mistakes
- Calling a cash position a short position.
- Assuming reversed performance must equal the negative of original performance.
- Changing thresholds or dates after seeing which direction wins.
- Ignoring different exposure, turnover, and incomplete final trades.
- Giving the reversed version more favorable cost or timing assumptions.
- Counting both directions as independent confirmation after choosing the better one.
Use reversal to try to disprove yourself
Reversing buy and sell rules is most valuable as a falsification tool. It forces the direction of the hypothesis to matter and reveals whether performance came from the claimed signal, general market exposure, or an implementation detail.
Define the reversed state before calculating it, keep the test long-only unless you deliberately model shorting, and report exposure and costs alongside return. If the reversal wins, call it a new exploratory finding. If neither version works, preserve the negative result.
Continue with parameter sensitivity and robustness, review overfitting in backtesting, or open TickRun to inspect a strategy’s original signals.