The direct answer, and the boundary we need to state first
Ask ten stock-signal services whether they beat the market and most will dodge the question, because answering it honestly requires publishing a benchmark-relative, dated record, not a curve of cumulative pips. We would rather state our own boundary upfront than let you find it three paragraphs in: Best Trading Signal publishes index and CFD signals — the S&P 500, NASDAQ and DAX/GER40, alongside gold, forex, oil and crypto. We do not publish single-company share picks. This is not a coy hedge; it is the single most important fact on this page, and it should not require reading past the first two paragraphs to find it. If what you want is an alert on a specific chipmaker or airline, this is not that service, and we would rather send you looking elsewhere than pretend otherwise.
That said, an index like the S&P 500 or NASDAQ is itself a basket of the same shares a single-stock service would trade — you get market-direction exposure through a CFD, with defined entries, take-profits and stop-losses, without picking one company's earnings call. If that fits what you actually need, our performance page has the full published record, losing weeks included. If you specifically need single-share alerts, keep reading: the evaluation checklist below applies to any provider you consider, including us on the index side. It is also worth noting up front that no index, no CFD and no equity ever moves in a straight line, so whichever coverage you pick, the question is never whether a losing week happens — it is whether the provider shows you that week honestly.
What 'beating the market' actually means
'Beating the market' is a specific claim, not a mood: it means a strategy's net return, after spreads, commissions and funding costs, exceeded a stated benchmark index over a stated period, adjusted for the risk taken to get there. A signal service that shows you a win rate or a raw points total without naming the benchmark and the period has not made that claim — it has shown you a number. A service that returned 8% while drawing down 40% along the way did not 'beat' a benchmark that returned 6% with a 10% drawdown, even though the headline figure looks bigger.
Most providers avoid this comparison entirely, and it is easy to see why: a benchmark forces an apples-to-apples test that a cherry-picked quarter of trades cannot survive. When you evaluate any service — including this one — ask three questions: which benchmark, over what dated period, and is the number by points or by trade count. Our own record is stated by points, not by trade count, and we say that every time we cite it, because the two measures are not interchangeable. If you are new to what a signal actually is before you get into evidence quality, start with what are trading signals.
The evidence hierarchy: from backtest to audited live record
Not all 'track records' carry the same weight, and the difference is not subtle. The table below ranks the evidence tiers you will encounter from a stock or index signal provider, weakest first. Understanding which tier a provider is actually showing you, versus which tier their marketing implies, is most of the work of evaluating any signal service, AI-powered or not.
Evidence tiers for signal-service track records, weakest to strongest
| Tier | What it is | Main risk |
|---|---|---|
| 1. In-sample backtest | Simulated on historical data the model was tuned on | Overfitting — fits noise, not a real edge |
| 2. Out-of-sample backtest | Tested on historical data held back from tuning | Still no live execution risk or slippage |
| 3. Forward-tested (demo) | Published in real time, before the outcome, on demo capital | Selection bias if losing runs get quietly dropped |
| 4. Live audited record | Real signals, dated publicly before the result, ideally third-party verified | Rare, but the only tier that is hard to fake |
Why backtested AI results are the weakest evidence
An 'AI-powered' backtest sounds like it should be more trustworthy than a human's, and it is usually the opposite. Machine-learning models are exceptionally good at finding patterns in historical data, including patterns that are pure noise and will never repeat. This is overfitting: the more parameters a model has and the more it is tuned against one historical dataset, the better it looks on that dataset and the worse it tends to perform on data it has not seen. The more sophisticated the model sounds in the marketing copy, the more scrutiny that specific claim deserves, not less. A backtest with a smooth, high equity curve and no explanation of out-of-sample testing is the single biggest red flag in this category.
Two more failure modes compound the problem. Look-ahead bias creeps in when a backtest accidentally uses information that would not have been available at the time — a next day's closing price folded into a signal generated that morning, for instance — inflating results in a way that is easy to miss from outside. Survivorship bias shows up when a provider tests a hundred variants of a strategy and only publishes the handful that happened to work, discarding the rest without disclosure. Ask any provider making AI or backtested claims how they controlled for both, and treat silence as an answer.