Technical analysis is the study of past price and volume to make trading decisions. It's also the most hype-infested territory in all of trading: "infallible" indicators, patterns "that work 80% of the time" (measured by nobody), courses selling esoteric geometry. This guide separates what the evidence supports from what the evidence demolishes — including my own evidence, published in full.
The organizing principle
An indicator is a compressible description of the recent past. It can measure the present; it contains no information about the future that price doesn't already have. Predictive value, if any, must be demonstrated net of costs — not asserted.
Support and resistance
The idea: levels where price reacted before tend to produce future reactions. The rational basis exists — real orders cluster around certain levels: stops, take profits, institutional orders — it's microstructure, not magic. The practical problem: eyeballed levels suffer confirmation bias (you notice the bounces, forget the breaks). Honest use: levels defined by repeatable mechanical rules (session highs/lows, impulse origin zones — as in my gold EA), never redrawn after the fact.
Moving averages
They describe the average trend over a window. As a regime filter ("longs only above the 200-period average") they're a simple, testable subtractive filter. As an entry signal (the "golden cross") they're late by construction and, tested net of costs on short timeframes, rarely survive. A technical note worth gold: any threshold based on absolute values must be expressed in relative units (ATR, percentages) — I personally paid for that lesson.
Oscillators: RSI, stochastic, MACD
They measure the relative speed of recent movement. "Overbought" does not mean "about to fall": in strong trends the RSI stays overbought for weeks while price keeps climbing. The adult use is as a measure of extension within context: how stretched is the move relative to its recent norm? The trading decision must come from the system, not from the magic 70/30 numbers.
Candlesticks and patterns
Hammer, engulfing, doji: a vocabulary for describing who won each bar's battle. As vocabulary, useful. As an isolated signal: I tested them systematically — long-shadow rejections, extended-body breakouts, on 210,240 Bitcoin bars, with leak-free features and an out-of-sample verdict. Gross edge ≈ zero, against 6.5 basis points of costs. And the conditional test ("but filtered by volume/volatility/session…") gave the same result: nothing. If someone claims otherwise, one question: where's the data?
Volume and VWAP
Volume is the most underrated information in retail: it says how much conviction sits behind a move. The VWAP — volume-weighted average price — is the benchmark institutions measure their executions against: around that level concentrate the decisions of people moving real size. It's the perfect example of an indicator used well: it objectively measures the present (where the session's value is, how far you are from it) without pretending to predict the future.
ATR: the indicator you should use regardless
The Average True Range measures recent average volatility. It doesn't generate signals: it generates units of measurement. Stops, targets and thresholds expressed in ATR multiples adapt automatically to the market — the difference between a system that ages well and one that breaks when the market changes scale. It's the unit of measurement in all my systems.
The test every signal must pass
- Formalization: the signal must be writable as an unambiguous mechanical rule. "Strong demand zone" is not a rule; "lower shadow ≥ 60% of the bar's range" is.
- Complete counting: all cases, not just the remembered ones. The chart a course shows you is a sample selected by its outcome.
- Baseline comparison: the signal must beat "always buy", not zero — on a rising asset, even nothing looks like it works.
- Costs in: realistic spread + commissions + slippage. This is where 90% of intraday signals die.
- Out-of-sample: verdict on data never used during development. The full protocol.
You have the tools to do this at home: the Python pipeline with free Binance data. Three hours of work worth more than any technical analysis course ever sold.