1. What an indicator actually is
A formula over past prices. A 20-period MA is the average of the last 20 closes - it can never tell you what the 21st will do. Indicators organise the past; they do not contain the future. Everything that follows depends on keeping that boundary sharp.
2. The three honest uses
Smoothing (moving averages): makes trend direction easier to see. Relative strength (RSI and kin): standardises momentum into a comparable number. Confirmation (volume): checks whether price moves have participation. Note what is missing: entry generators. Indicators inform context; the plan's rules make decisions.
3. The parameter-fitting trap
Every extra indicator, period or threshold is a dial you can turn until history looks perfect. Ten dials can "explain" anything. That is curve fitting: a backtest with many free parameters is a description of the past wearing a prediction costume. The honest test is out-of-sample data the dial-turner never saw.
Remember: If your strategy needs RSI 30/70 on a 14-period with a 2% band on a 4-hour chart, you have not found an edge - you have tuned ten dials to one asset's history.
4. The fees-and-slippage illusion
High-frequency indicator strategies (scalping, cross-signal scalping) look brilliant in backtests and die in execution: 0.1% taker fees each way plus slippage consume small edges. A 95% win rate with tiny wins and huge tail losses is the classic printed-backtest shape - and the classic account-killing one.
5. When this fails
Indicator logic fails silently in three ways: regime change (the correlation it tracked ends), crowded trades (everyone sees the same signal, liquidity vanishes on exit), and data issues (fake volume feeds the input). Any backtest result is a historical anecdote until re-verified out-of-sample.
Risk: Signals that arrive in screenshots, Telegram channels or "AI indicators" with claimed win rates have no place here. Unverifiable claims are the marketing layer of manipulative products.