A trading method can use indicator values to define decisions. Processing earlier prices does not prove future profitability or prove that profitability is impossible. An indicator defines a calculation; a testable strategy also needs data, entry, exit, sizing, execution and costs. Assess that complete rule rather than assign an edge to a line’s appearance.

What an indicator actually is and where its numbers come from

Check an indicator’s declared inputs and formula. SMA sums selected observations and divides by their count; for hypothetical EUR/USD closes 1.1000, 1.1010 and 1.1020, SMA(3) = 1.1010. That is not a forecast of the next close. Changing the price source or window changes the calculation. Some scripts use additional data, so do not assume every indicator uses only prices visible on this chart.

A price transformation does not supply an independent observation just because it has a new name. It may still make a rule easier to define and reproduce. For a signal such as close crossing above an SMA, state the instrument, timeframe, confirmation time, available entry price, exit, size and treatment of another signal during an open trade. “Buy when the line rises” is not a complete reproducible test.

Why indicators are derived and lagging

A historical calculation and a forecast are different objects. A method can use a current indicator value as a condition, but it must test the proposed future outcome separately. Oscillator extremes do not themselves identify exhaustion or a reversal time. Current-bar values can change before close, and a later confirmed value must not be treated as if it was already known earlier.

Several indicators may process the same inputs, so three agreeing colours are not automatically three independent confirmations. Measure the contribution of an additional rule in a separate evaluation rather than impose a compulsory one-trend/one-volatility/one-level set. The article on leading versus lagging indicators provides related terminology; it does not establish a guaranteed lead for a chart oscillator.

Same result, different drawdown

Arrange the same four results in two orders. Profitability comparisons should also include the path from a peak to a later trough.

Scroll sideways to see the full diagram

Same result, different drawdownModel A: +2, −1, +2, −1 R, cumulative 0, 2, 1, 3, 2. B: +2, +2, −1, −1 R, cumulative 0, 2, 4, 3, 2. Both win two of four and finish +2 R. Maximum close-to-close drawdown is 1 R in A and 2 R in B.Total and win rate do not show loss order0 R1 R2 R3 R4 R01234Number of closed tradesA: +2, −1, +2, −1 R · max DD 1 RB: +2, +2, −1, −1 R · max DD 2 RBoth: total +2 R · win rate 50%The same average does not mean the same path.

Hypothetical model: R is a fixed reference amount, results are already net, with no reinvestment or open positions. We measure drawdown of cumulative closed results in R, not equity percentage or all future strategy risk. Peak-to-following-trough definition: MetaTrader 5 report.

Trade order changes drawdown even with an equal total.

What really decides profitability

Let p be the share of wins, W the average gross win, L the average gross loss as a positive amount and c the average cost. With no zero-result trades, average net result is p × W − (1 − p) × L − c. For historical data, this is a sample description, not a known expectation of future trades. Win rate alone cannot determine the sign; costs and loss sizes matter.

Practical takeaway: Assess win rate together with payoffs, costs and loss order. A high win percentage alone does not establish positive expectancy.

In a fully hypothetical A model of 100 closed trades, 40 gross wins of USD 200 and 60 gross losses of USD 100 give USD 2,000 before costs. A fixed all-in cost of USD 25 per trade subtracts USD 2,500: net total −USD 500, mean −USD 5. In B, 70 wins of USD 50 and 30 losses of USD 150 give −USD 1,000 gross; USD 10 cost per trade gives −USD 2,000 net, mean −USD 20. These are invented outcomes and costs, without reinvestment or a test of any indicator.

A’s conditional break-even win share is (100 + 25)/(200 + 100) ≈ 41.67%. Changing exits can change p, W and L together; do not keep an old win share fixed while inventing a larger target. Use complete net history, and do not subtract spread again when actual execution prices already include it. The expectancy formula explains the calculation; RSI and MACD explain particular indicator inputs, without certifying an edge.

Availability, costs and parameter selection

Choosing the best configuration from many historical trials can fit noise in that sample. Carr and López de Prado discuss this calibration problem; their proposed model concerns prices following a discrete Ornstein–Uhlenbeck process, not proof that your indicator works. Keep all tested variants and distinguish the development sample from a later unused evaluation. A sharp deterioration does not alone identify the cause: costs, data errors, changed conditions and randomness also need investigation.

A confirmed chart can differ from what an indicator showed during an open bar. Record the signal time, inputs and subsequent available fill price and compare that record with the final chart. Normal live updates are not automatically fraudulent repainting. For a paid tool, review disclosed methodology, complete results and subscription costs; price alone proves neither quality nor its absence. The separate guide to paid forex signals develops that assessment.

The next comparison separates average return per accepted signal from total return over the same period. A higher average alone cannot tell you whether a new filter improved the method’s drawdown or total outcome.

A better average can accompany a smaller total

Use 20 hypothetical non-overlapping opportunities over one common period, with a fixed reference amount R, results already net of assumed costs and no increase in size. A filter retains four results [2, 2, −1, 1] R and rejects eleven +1 R and five −1 R results. This is an invented arithmetic set, not an empirical indicator study.

All opportunities 20 observations: total 10 R, average 0.50 R. Retained: 4 observations, total 4 R, average 1.00 R.
Rejected opportunities 16 observations: total 6 R, average 0.375 R. The filter doubled average per accepted trade but reduced the same-period total from 10 R to 4 R.

The filter might also change drawdown, time committed or another constraint, but those effects need measurement rather than inference from a higher mean. Increasing size would change this comparison’s assumptions. Report count, average and total in the same units and period. Proportionately scaling all wins, losses and costs preserves the sign of their mean; fixed minimum fees can break that proportionality. Risk control does not turn a losing rule into a profitable one. The guide to discovering a trading edge discusses testing a complete method rather than presume an outcome from a plan.

What to do tomorrow

  1. Freeze the rules. Record instrument, feed, timeframe, parameters, entry, exit, size and costs in one version before assessing it.
  2. Keep the sequence. Report all net results, average, trade count and drawdown, plus rejected or unfilled orders. A month’s sample mean is not a guaranteed future expectation.
  3. Use available information. Reproduce the signal’s real confirmation time and fill assumptions; do not insert later prices into an earlier decision.
  4. Evaluate unused data. Separate development from a later period without retuning. Changing rules after seeing that evaluation consumes it as an untouched test.
  5. Evaluate a paid tool’s evidence. Review the full history, methodology and annual cost rather than select only winners. The trading strategies section on ForexMechanics is further reading. Demo or paper work can reveal problems without requiring you to risk money, but does not guarantee future profit.