Automated technical analysis: what can (and can't) be automated

The short version

Indicators: fully automatable

Moving averages, RSI, average true range, VWAP and their relatives are formulas. Given the same candles and parameters, software computes them exactly and instantly. The risk isn't the calculation; it is the choices around it. A 14-period RSI and a 9-period RSI disagree, and a gap in the data distorts every average that includes it. Automating indicators makes those choices explicit and consistent, which helps only if you made them deliberately.

Chart patterns: mostly automatable

Patterns are geometry plus rules, so they automate well once you accept a precise definition. The cost is that every "roughly equal" or "clearly converging" in the textbook becomes a number, and every number excludes some shapes a person would accept. The mechanics are covered in automated chart pattern detection, and running them across many markets in chart pattern scanners.

Multi-timeframe checks: mostly automatable

Software can check whether a pattern on the 15-minute chart agrees with the trend on the 1-hour chart, but it has to be told how to treat disagreement. Requiring every timeframe to agree removes many signals; ignoring higher timeframes throws away useful context. A common compromise, and the one PatternX uses, is to let the higher timeframe raise or lower a score rather than veto the trade.

Risk management: the rules automate, the judgement doesn't

Risk rules are where automation helps most, because they are the hardest rules to follow under pressure. Software can size every position from a fixed risk per trade, enforce a daily loss limit, cap open positions and pause after a losing streak without negotiation.

For example, with a $10,000 balance and 1% risk per trade, you are risking $100. If the stop is 2.5% below entry, the position is $100 ÷ 0.025 = $4,000. Software does this arithmetic perfectly. What it can't decide is whether 1% is right for you, whether to use leverage, or whether this is a period to trade at all. Those inputs are yours.

Execution: automatable, with safeguards

Placing an order is easy to automate. The hard parts are what happens around it: a stop the venue rejects, a connection that drops after the entry fills, a price that gaps through the stop. A well-built system places the protective stop-loss at the broker or exchange right after the fill, so it rests at the venue even if the software or your connection fails, and has a defined response if the venue refuses it.

What stays human

• Context. Whether the market is in a phase where breakouts follow through, or one where they routinely fail, is only partly visible in price data.
• News judgement. An earnings release, a token unlock, an exchange outage or a regulatory announcement can override any chart. Price-based software can't read them, and treating a language model's news summary as a trading signal brings its own risks.
• Position sizing decisions. The formula is mechanical; the inputs (risk per trade, leverage, how much capital to commit) are judgements about your own situation.
• When to switch it off. Recognising that your rules no longer fit the market is a human call.

Pitfalls

Overfitting

If you test many variations on the same history and keep the winner, part of what you kept is luck. For example, try 12 combinations of take-profit, stop-loss and minimum confidence: even if none has a real edge, one will look strongest. The defence is to judge it on data it wasn't chosen on, an out-of-sample period, and to distrust results built on few trades.

Curve fitting

Curve fitting is the related habit of adding conditions until every historical loser disappears. A rule such as "only trade flags on Tuesdays when RSI is between 52 and 58" can explain the past perfectly and predict nothing. Each condition should have a reason you could state before seeing the results.

Ignoring costs

Small per-trade edges vanish after costs. For example, a strategy averaging +0.25% per trade before costs, paying 0.05% in fees and losing 0.03% to slippage on each side, keeps 0.25 − 0.10 − 0.06 = 0.09% per trade, about a third of its gross edge. On leveraged perpetual futures, funding payments add another cost. A backtest that skips costs describes a market you can't actually trade.

How PatternX automates it

PatternX automates detection, scoring, risk checks and execution, and leaves the switches with you.

• Manual or automatic. Each exchange connection runs in manual mode by default: signals wait for your approval in the app or via Telegram and expire after 15 minutes. In automatic mode they execute without approval but still pass every risk check.
• Paper or live. Connections start in paper trading mode with a $10,000 simulated balance, fills at live bid and ask prices and fees estimated at 0.05% per side. Live trading requires the platform's live-trading switch, a connection that has passed its test and an API key that isn't known to allow withdrawals.
• Risk engine. Before every order it checks kill switches, daily trade quotas and, with default settings, a daily loss limit of 10% of balance, at most three open crypto and three open traditional trades, a 60-minute cooldown after three losses in a row and a US pattern-day-trader guard. Positions are sized from 1% risk per trade by default. See risk controls.
• Stop at the venue. On live trades the stop is placed at the exchange or broker right after the fill, using the venue's own mechanism. If the venue refuses it, PatternX closes the position at market and alerts you rather than leave it unprotected. Break-even and trailing-stop adjustments are managed by PatternX's software loop, not changed at the venue, so they work only while PatternX is running and can reach the venue.
• Testing. The backtester replays the same engine bar by bar with fees, slippage and funding, and splits results into in-sample and out-of-sample periods. It doesn't model the risk engine, trailing stops or approval delays.

A sensible order of steps

1. Backtest with costs, and read the out-of-sample period first.
2. Paper trade with manual approval, so you see every signal.
3. Go live with manual approval and small size.
4. Switch to automatic execution only for setups whose behaviour you have watched.

Trading carries a risk of loss at every step. Automation changes how decisions are made, not whether trades can lose.