Trading statistics terms

Demo account results

Also called: Demo results.

Demo account results are performance figures produced on simulated money, and they systematically overstate what the same strategy achieves live because the costs that hurt most are the ones a demo does not charge.

A demo server fills orders against quoted prices with unlimited liquidity and no market impact. Live execution supplies none of that. The differences concentrate exactly where short-term strategies earn their margin: slippage on entry and exit, spread widening during news and rollover, requotes, and rejections that a demo simply never issues.

The gap is largest for the strategies most often sold. A scalper taking small targets can have its entire edge consumed by execution quality; a position strategy holding for days barely notices. So the demo-to-live drop is not uniform noise — it is a systematic penalty weighted toward exactly the products marketed with the most impressive demo curves.

Treat demo results as evidence that the software runs, not that the strategy works. They confirm the logic executes, the orders are formed correctly and nothing crashes overnight, which is worth knowing. They do not establish profitability, and a vendor presenting demo performance as a track record has answered a question nobody asked.

Deposit load

Also called: Deposit load percentage.

Deposit load is the share of account equity committed as margin at the busiest moment in the record, and it is the fastest way to see how much risk produced a published return.

Two accounts can post the same return having taken entirely different risks. One committed 3% of equity as margin at its peak; the other committed 60% and survived. The return column cannot distinguish them. Deposit load can, which is why it is usually reported somewhere small.

High load means positions large enough that an ordinary adverse move reaches the margin call level. It is the statistical fingerprint of martingale and grid systems, which accumulate positions against a losing move and therefore look flawless until the one move that does not come back.

Read it together with maximum drawdown, because the pair tells you what the equity curve alone hides. Low drawdown with high deposit load is the dangerous combination: it means the account has been carrying large exposure and has simply not yet met the move that punishes it. A smooth curve is not evidence of safety when the load is high — it is evidence that nothing has gone wrong yet.

Expectancy

Expectancy is the average amount a strategy is expected to win or lose per trade, calculated from win rate, average win and average loss.

The formula is (win rate × average win) minus (loss rate × average loss). Expressed in R multiples it becomes portable across account sizes: an expectancy of 0.15R means each trade is worth 0.15 times the amount risked, on average.

This is the number that determines whether a system is worth running, and it combines the two figures that are meaningless separately. A 35% win rate with 3:1 payoff gives 0.4R; a 60% win rate with 0.5:1 payoff gives −0.1R, and that second system loses money while winning most of its trades.

Costs must be inside the calculation, not added afterwards. On short-target gold strategies spread and commission routinely consume a double-digit share of the gross move, which is enough to move a marginal expectancy below zero — the profit calculator shows that share for a specific trade.

Why even a profitable record can be luck, and how often it is, is worked out in is forex trading profitable?

Gain percentage

Also called: Gain, Total gain.

Gain is the headline return figure on a published track record, calculated so that deposits and withdrawals do not count as performance — which makes it more honest than a raw balance change and easier to misread.

A raw balance change is meaningless on an account that has been topped up: deposit $5,000 into a $5,000 account and the balance doubles without a single trade. Gain corrects for this by measuring only the part attributable to trading, typically by chaining the returns of each period between cash movements.

The correction introduces its own distortion. Because the periods are compounded, a large gain earned while the account was small keeps its full weight forever, even after the account grows tenfold and never repeats it. An account showing 400% gain may have made almost all of it in one early month on a few hundred dollars — the figure is not wrong, it simply is not what a new subscriber will experience.

The reliable reading is to ignore the headline and open the monthly table. Twelve mediocre positive months are a system; one enormous month and eleven flat ones are an event. Cross-check against maximum drawdown and deposit load, because a high gain produced at high risk is a description of the risk, not of the skill.

Profit factor

Profit factor is gross profit divided by gross loss, so a value above 1.0 means a strategy made more than it lost over the measured period.

It is a better single summary than win rate because it incorporates the size of wins and losses rather than only their count. Roughly, values below 1.2 are fragile, 1.3 to 1.6 is a realistic band for a genuine retail edge, and anything above 2.0 over a short sample should be assumed to be curve-fitted until demonstrated otherwise.

The metric is highly sensitive to outliers on small samples. One exceptional winner can lift profit factor from 1.1 to 1.8, which describes luck rather than edge. Checking the figure with the single best trade removed is a fast and revealing sanity test.

It also ignores sequence entirely. A profit factor of 1.5 achieved by a smooth climb and one achieved by a 40% drawdown followed by a recovery are indistinguishable in this measure, which is why it should always be read alongside maximum drawdown.

R multiple

An R multiple expresses a trade's result as a multiple of the amount risked, so a trade that gained twice its stop distance is +2R regardless of account size.

Normalising results to R removes account size, position size and instrument from the comparison, which makes trades from different periods and different symbols directly comparable. It is the standard unit for evaluating a strategy rather than an account.

Recording every trade in R also removes the emotional distortion of dollar amounts. A −1R loss is a −1R loss whether it cost $12 or $1,200, and a system judged in R is judged on its behaviour rather than on how it felt.

The requirement is that R is defined at entry and never adjusted afterwards. Widening a stop mid-trade and then reporting the result against the original risk understates the loss, and it is the most common way trade journals quietly become fiction.

Sample size

Sample size is the number of trades a performance figure is computed from, and it determines how much confidence any of those figures deserve.

Thirty trades is enough to notice a very large effect and nothing else. A hundred begins to be informative. Several hundred is where win rate and profit factor stabilise sufficiently to be treated as estimates rather than anecdotes.

This is a genuine and unavoidable problem for low-frequency strategies. A system taking ten trades a month generates 120 a year, so demonstrating an edge with reasonable confidence takes years, not months. Nothing can compress that timeline — running the same period through more backtests does not create more independent observations.

It is why our own figures carry the sample alongside them and why we say plainly that roughly ten trades a month makes any percentage unreliable. A vendor quoting a precise win rate from a few weeks of live trading is either unaware of this or relying on the reader being unaware.

How many trades it takes before a record means anything is worked out in how many trades before a track record means anything?

Sharpe ratio

The Sharpe ratio measures excess return per unit of volatility, allowing strategies with different risk levels to be compared on a common basis.

It divides return above the risk-free rate by the standard deviation of returns. The intent is to reward consistency: a strategy returning 15% with small fluctuations scores better than one returning 20% with violent ones.

Its main flaw for trading systems is that it penalises upside volatility identically to downside. A strategy with occasional very large winners is punished for exactly the behaviour that makes it valuable, which is why the Sortino ratio — using only downside deviation — is often preferred.

It is also unstable on small samples. Ten trades a month for seven months is roughly seventy observations, which is nowhere near enough for a Sharpe ratio to mean anything. We do not publish one for that reason, and any retail product quoting a precise Sharpe over a short live period is quoting noise.

Standard deviation

Standard deviation measures how widely values are dispersed around their mean, and in trading it is the most common quantification of volatility.

Roughly two thirds of observations fall within one standard deviation of the mean and about 95% within two, provided the distribution is normal. That proviso is doing heavy lifting, because financial returns are not normally distributed.

Real markets have fat tails: extreme moves occur far more often than a normal distribution predicts. A move described as a five-standard-deviation event should happen roughly once in several thousand years and in practice happens every few years, which means models built on normality systematically understate tail risk.

It is nonetheless the basis of Bollinger Bands and of most volatility scaling, and it works acceptably for the ordinary middle of the distribution. The mistake is trusting it about the extremes — which are exactly the events that end accounts.

Statistical significance

Statistical significance is the judgement that an observed result is unlikely to have arisen by chance alone, given the sample size and the size of the effect.

In trading the question is whether a strategy has an edge or merely had a good run. With enough strategies tested against enough data, some will look excellent purely by chance — this is not a hypothetical, it is arithmetic, and it is why testing many ideas requires more evidence per idea, not less.

The standard defence is to hold data back. Fit and refine on one period, then test once on a period never used during development. A result that survives that is not proof, but it is dramatically better evidence than a result that has never left the data it was born in.

We ran exactly this check. The shipping configuration was developed on January to July 2026 and then tested on August to December 2025, which it had never seen. It returned +$154 and was positive in three months of five — profitable, but not the "green every month" the in-sample period suggested. That gap is what out-of-sample testing exists to reveal.

What it takes to tell a real edge from luck in a trading record is worked out in how many trades before a track record means anything?

Wilson score interval

The Wilson score interval is a confidence interval for a proportion that behaves sensibly on small samples, unlike the naive normal approximation.

The problem it solves is specific. If a strategy has won three of four trades, its observed win rate is 75%, and treating that as the true rate is obviously wrong. The Wilson interval gives a range — in that case roughly 30% to 95% — that honestly reflects how little four trades tell you.

Its lower bound is the useful quantity for ranking. It answers "what is the worst the true win rate plausibly is, given what we have seen", which naturally penalises strategies with few observations without excluding them entirely.

This is the mechanism behind our strategy tournament. Each strategy is scored on the lower bound of its measured win rate rather than the raw figure, so a strategy that won its only two trades does not outrank one with a long and solid record. It is the single most important reason the engine does not chase whichever strategy got lucky most recently.

Win rate

Also called: Hit rate, Strike rate, Calculate win rate, How to calculate win rate.

Win rate is the percentage of trades that close profitably, and on its own it says nothing useful about whether a strategy makes money.

A 90% win rate is achievable trivially by taking tiny profits and letting losers run, and it describes a strategy that will eventually be destroyed by one trade. A 35% win rate with a 3:1 payoff is a strong system. The number is meaningless without the payoff ratio beside it.

It is nonetheless the number most heavily marketed, because it is the most intuitive and the easiest to make look impressive. Any product advertising a win rate without a corresponding average win and average loss is selecting the flattering half of the picture.

Our validated win rates are 36% in-sample and 37% out-of-sample, at a fixed 2:1 reward-to-risk. They are effectively the same, which is a better result than a fall would have been — the in-sample number benefits from the same data the system was refined on, so a gap would have measured fitting. What did not hold steady is the return: the out-of-sample period made three times as much, which is the number to distrust.

The win rate each risk-reward ratio needs, and what gold actually delivered, is in does the risk-reward ratio actually work?