What a Proxy Backtest Can (and Can’t) Tell You
Not every backtest can reconstruct the original strategy, market, universe, or data source exactly. Sometimes the honest test is a proxy: a defined substitute that can be tested consistently, with a conclusion that stays narrower than the original strategy claim.
That doesn’t make the result weak by default. A proxy is a defined substitution. The problem is not using a proxy. The problem is hiding the substitution or letting the conclusion grow beyond what the substitute can support.
Key takeaways:
- A proxy backtest replaces part of the original setting with a substitute that can be tested consistently.
- The result is useful only when the substituted piece is disclosed and the conclusion is limited to the tested implementation.
- Benchmark choice determines the question the test is actually answering.
A proxy result is evidence about the tested substitute implementation; it is not automatically evidence about the original strategy’s native market, universe, or execution context.
What is a proxy backtest?
A proxy backtest replaces part of the original strategy setting with a clearly defined substitute. The substitute might be a different instrument, a narrower universe, an available data series, or a deterministic implementation of a rule that was originally described in less testable terms.
The point is not to pretend the proxy is the original thing. The point is to answer a narrower question cleanly. Instead of saying, “this is what the original strategy did,” a proxy test should say, “under this substitute implementation, with these assumptions, this is what the rule produced.”
Why proxy tests are used
Proxy tests are useful when the original strategy idea is clear enough to study, but the original setting cannot be reconstructed with enough confidence. That can happen when:
- The original market data is unavailable or outside the available test data.
- The original universe cannot be reconstructed consistently.
- The source rule is underspecified, so a deterministic implementation must be chosen.
- A tradable substitute is available, while the original exposure is not practical to test directly.
- The research question is about one active decision, not a full replication of the original ecosystem.
Used carefully, a proxy can preserve the spirit of a research question while making the test observable, repeatable, and easier to compare under a standardized backtesting methodology.
What exactly was substituted?
The first question is not whether the proxy result was good or bad. The first question is what changed between the original idea and the tested implementation.
- Instrument proxy: using one tradable instrument as a substitute for a broader or less accessible exposure.
- Universe proxy: using a smaller, cleaner, or more accessible universe instead of the original universe.
- Data proxy: using an available data frequency or source when the original rule may have implied different data.
- Implementation proxy: translating an underspecified rule into a deterministic set of testable assumptions.
Benchmark choice is related, but it is not the same kind of substitution. The benchmark is usually the comparison frame, not the thing being proxied. It determines whether the test is asking about the value of an active rule, the opportunity cost of an allocation, or the behavior of a broader opportunity set.
What a proxy backtest can tell you
A proxy backtest can show how a rule behaved under the substitute implementation. It can show whether the rule helped or hurt the tested path. It can also show whether practical burdens such as timing, costs, turnover, short exposure, drawdown, or benchmark opportunity cost shaped the result.
It can also show whether a question deserves deeper follow-up research. That’s different from saying the proxy has settled the original question. The value is in identifying what the tested rule did under known assumptions.
What a proxy backtest cannot prove
A proxy backtest cannot automatically prove that the original strategy would have worked or failed. It cannot make a single-instrument test equivalent to a native-universe result. It cannot make an implementation choice speak for the source strategy. And it cannot safely generalize one substituted test across an entire strategy family.
The Donchian Four-Week Rule backtest is a useful example. A SPY proxy implementation can test what happened when a Donchian-style rule was applied to one equity-index ETF. It is not a full diversified futures replication, and it should not be read as a verdict on all Donchian-style trend following.
Benchmark choice determines the question
The benchmark must match the proxy question. A same-instrument benchmark can help isolate whether the active rule improved the path of the tested instrument. A broader benchmark may be needed when the question is about allocation, opportunity set, or original-universe behavior.
That’s why choosing the right benchmark is part of interpreting the proxy itself. The wrong comparison can make a narrow test look broader than it is, or make a useful proxy look irrelevant because it is being judged against the wrong question.
How to read proxy results
Before reading the performance result, read the boundary around the test. A proxy result is most useful when the page makes that boundary visible.
- What was substituted?
- Why was the substitute used?
- What stayed the same?
- What benchmark matches the actual question?
- What conclusion is now narrower than the original strategy claim?
A proxy has the same implementation problem as a signal. A signal may be interesting, but it still needs rules, assumptions, risk handling, costs, and a comparison frame before the result can be interpreted. The same discipline applies when the test uses a proxy. See A Signal Isn’t a Strategy for the related implementation problem.
Red flags in proxy backtests
Proxy tests become misleading when the substitution is hidden or the conclusion expands beyond the tested setup. Common warning signs include:
- The page does not disclose that a proxy was used.
- The benchmark does not match the proxy question.
- The copy claims original-universe replication without original-universe data.
- The result is generalized to an entire strategy family.
- The proxy changes multiple major assumptions at once without explaining which change matters.
- The page reports performance without saying what the substitute instrument, universe, data, or implementation was.
Bottom line
A proxy backtest is not a loophole. It is a narrower experiment. When the substitute is disclosed and the benchmark matches the question, the result can be useful. When the substitute is hidden or the conclusion is overgeneralized, the result becomes misleading.
The best proxy pages are modest. They do not ask the reader to treat a substitute as the original thing. They show exactly what was tested, explain why that substitute was used, and keep the conclusion inside the evidence.