Trifecta Signals

How to Read a Track Record

Every signal service publishes numbers, and the numbers always look good — that's the point of publishing them. Reading a track record well means asking how each figure was produced: what was counted, from what starting point, over how many cases, and what was left out. This guide walks through the questions that separate a record you can evaluate from one that's mainly marketing. The examples below use made-up numbers to show the method; they describe no real service.

Start with the measurement basis

A signal usually has more than one meaningful date. There's the day the pattern first appeared — the early signal — and, for methods that use one, the later day a follow-up condition was met — the confirmation. Returns measured from the early date and returns measured from the confirmation date describe different things, and they are rarely similar.

Here's why. A confirmation, by definition, arrives after some evidence has accumulated. Measuring from the confirmation date starts the clock later, often after part of the move has already happened. Measuring from the early date captures the full arc, including the uncomfortable stretch before anything was confirmed. Neither basis is dishonest. But a hit rate quoted without its basis is an incomplete claim, the way a batting average would be if you didn't know whether walks counted as at-bats.

When you compare two services — or two statistics from the same service — first check that the starting points match. A service reporting results from the early signal date and one reporting from confirmation are answering different questions. The more useful record says, plainly, which date every return is measured from, and applies that rule to every row.

What a hit rate does and doesn't tell you

A hit rate is the share of signals that reached some defined outcome — say, a gain of at least 10% within a fixed period after the signal. Imagine a hypothetical service that reports a 72% hit rate. That sounds strong until you ask three follow-ups.

First, what was the target and the window? Reaching +10% within 90 days is a different achievement than reaching +10% "eventually." A defined window matters because an open-ended claim can be true of almost any signal given enough time. Second, how were the signals that fell counted? A complete record keeps the losers in the denominator — including signals on companies that were later delisted. Third, what happened on the way? A signal that dropped 25% before recovering to +10% and one that rose steadily both count as hits. The hit rate can't tell them apart; the drawdown figures can, which is why they belong next to it.

Sample size: the quiet qualifier

Percentages built on small samples swing wildly. Suppose a made-up service shows a 90% hit rate — across 10 signals. One more loser drops it to about 82%. The same percentage across 1,000 signals would barely move. As a rough habit: treat any rate built on fewer than a few dozen cases as anecdote, a few hundred as suggestive, and a few thousand as a stable description of that method's past behavior — still not a promise about its future.

Sample size has a second dimension: time. A record spanning one strong market year describes performance in one kind of weather. A record spanning several years, including flat and falling markets, tells you how the method behaved when conditions turned against it. Ask what market environments the record covers, not just how long the list is.

Drawdowns belong next to every return

A return figure tells you where a signal ended up. A drawdown tells you the worst dip along the way — how far below the starting price the position fell before the measurement date. Two methods can post identical average returns while one routinely dips 8% first and the other routinely dips 30%. For anyone actually following the signals, those are very different experiences, and the second is far harder to stick with.

Useful records report drawdown honestly: a typical (median) figure and a worst case, not just an average that blends the two. If a record shows returns without any drawdown information, treat the picture as unfinished and ask for it. The concept is covered in depth in Drawdowns and Position Sizing.

Dedupe, overlaps, and double counting

Companies can signal more than once, and related share classes of the same issuer can signal at nearly the same time. If every repeat counts as a fresh, independent result, a record can quietly inflate both its sample size and its hit rate — the same underlying event wearing two jerseys. Ask whether the record dedupes by issuer: whether repeat signals for the same company within a short span are treated as one event for statistical purposes. A careful record says so explicitly. The same question applies to statistics quoted in marketing: headline figures should come from the deduped set, not the raw list.

Survivorship and the missing names

The most flattering edit a record can make is invisible: dropping the failures. Companies get acquired, merge, or delist, and a record rebuilt from today's list of existing companies silently loses every name that didn't survive. That's survivorship bias, and it flatters any method because delisted names skew heavily toward bad outcomes. A trustworthy record keeps delisted and acquired names in place, badged as what they are, with their results intact. If a service's history contains only companies that still trade today, its past has been tidied — and its percentages should be discounted accordingly.

Questions to ask any signal service

Before crediting any published record, you should be able to answer these from the service's own pages:

A service that answers these questions clearly is showing its work. One that leads with a single large percentage and answers none of them is asking to be trusted on faith. The vocabulary used across this guide — signal, confirmation, hit rate, drawdown — is defined in the Glossary, and the reasoning behind watching price action at all is laid out in Technical Analysis, Explained.

Educational material only — not investment advice.