If It Isn’t Backtested, Stop Using It
If an indicator hasn’t been reliably backtested—or you haven’t reviewed the results—stop using it. Set it aside until it earns a place in your process.
Trading content is full of polished demonstrations. An indicator flashes. The market turns. A line crosses, a bar changes color, or a signal appears just before a major move. The chart looks convincing because someone selected the example for that effect.
That can help explain how an indicator is calculated or what a signal looks like. It doesn’t show whether the indicator is reliable.
Given enough market history, I could probably find a few days when I scratched my foot and the market rallied afterward. The coincidence would be real. The indicator would still be absurd.
An indicator is a hypothesis
An indicator can be mathematically precise and still have no useful trading value. A formula tells us how the indicator is calculated. It doesn’t tell us whether the signal arrives early enough, works often enough, survives different markets, or produces an outcome worth acting on.
Until those questions are tested, the indicator is a hypothesis. It may deserve research, but it doesn’t yet deserve capital.
This distinction matters because a chart can look plausible without providing evidence. A moving average can appear to frame a trend. An oscillator can seem to identify a reversal. A colored bar can line up beautifully with a breakout. None of that shows what happened across the full record.
Examples can explain. They cannot validate.
A successful chart example answers one narrow question: Can I find a time when this signal lined up with a market move?
For almost any flexible indicator, the answer is yes. Markets produce thousands of turning points, breakouts, reversals, trends, consolidations and false starts. Give someone enough securities, dates, parameters and chart windows, and an impressive example will usually appear.
The harder question is the one that matters: What happened across every valid occurrence of the signal under one declared rule set?
A highlight reel can’t answer that question.
The missing denominator
Two successful examples tell us almost nothing without the denominator.
Were those two winners drawn from three signals, 30 signals or 300? How many signals failed? How many arrived late? How many produced a small gain before a larger loss? How often did the indicator fire during sideways markets, volatile markets or long declines?
Without the full sample, there is no signal count, no average or median return, no return distribution and no way to see whether a few memorable successes are carrying an otherwise weak record.
Once the signal is translated into a strategy, the backtest also needs to show the standard scorecard: CAGR, volatility, max drawdown, Sharpe ratio, Calmar ratio, ending capital and benchmark trade-offs.
The denominator is what separates an attractive anecdote from testable evidence.
If you haven’t reviewed the test, you don’t know the indicator
Even the phrase “backtested indicator” isn’t enough. A backtest can be vague, selective or built on assumptions that flatter the signal.
You need to know exactly what was tested: the traded universe, signal formula and parameters, entry and exit rules, decision and execution timing, holding period, position sizing, rebalance cadence, cost model, portfolio accounting and primary benchmark. The test also has to include every valid signal, not just the examples that support the story.
What Makes a Backtest Reliable? explains why the strategy rules and the testing environment both have to be clear. A Signal Isn’t a Strategy explains the additional rules needed before an indicator becomes a complete, testable process.
If those details are missing, the indicator may still be interesting but it has not earned trust.
The full record is the evidence
A useful test should make the full record visible, including the parts that don’t look good in a video thumbnail.
- All valid signals: not a hand-picked set of winners.
- Signal-level evidence: signal count, average and median return, and the full distribution of outcomes.
- Strategy performance: CAGR, volatility, max drawdown, Sharpe ratio, Calmar ratio, ending capital and benchmark comparisons.
- Trading profile: time in market, trades per year, win rate, turnover, costs and liquidity once the signal is translated into portfolio rules.
- Market context: whether the signal behaves differently across trends, declines, volatility regimes or sideways periods.
The point isn’t to demand a perfect result. Weak, mixed or regime-dependent results can all be useful. What isn’t useful is pretending that two successful examples describe the whole distribution.
A useful indicator has to survive ordinary markets
The most persuasive demonstrations usually feature dramatic moves: the top before a crash, the bottom before a rally, or the breakout before a major trend. Those are the moments everyone wants an indicator to capture.
Most trading days aren’t dramatic. Markets drift, chop, reverse, stall and produce signals that go nowhere. A useful indicator has to survive that ordinary history too.
That’s why the boring parts of a backtest matter. They reveal whether the signal has repeatable information or merely a few spectacular anecdotes.
Stop promoting hypotheses into live rules
The operating rule is simple: if an indicator hasn’t been reliably backtested, don’t treat it as a trading input. If a backtest exists but you haven’t reviewed the rules and results, don’t assume the test supports the claim.
No reliable backtest. No place in the process.
This isn’t hostility toward indicators. They’re useful research tools that can organize information, define conditions and suggest questions worth testing. But no indicator should be promoted from an idea to a live rule because someone found two charts where it looked brilliant.
Real money should not be the first serious test.
Backtesting doesn’t create certainty
A credible backtest does not prove that an indicator will work in the future. It does not guarantee live execution, eliminate data mining or turn historical relationships into permanent laws.
What it can do is replace a selected story with a defined body of evidence. It shows what happened when the same rules were applied repeatedly, including the failures, frictions and uncomfortable periods.
That’s a lower claim than certainty. It’s also a far higher standard than a chart with two arrows on it.
For the broader testing standard, read A Backtest Should Be a Proving Ground, Not a Sales Pitch.
An example earns attention. A test earns consideration.
A chart example can make an indicator worth investigating. It can’t make the indicator reliable.
The standard is simple. Define the signal. Test every valid occurrence. Specify the execution, sizing, cost and benchmark rules. Review the full result. Then decide whether the indicator deserves further research.
Until then, stop using it.