A client sent me their trading strategy for automation. At first glance, solid. Clear areas of interest, logical market approach. The problem? Half the rules were in their head. "I'll wait for confirmation," "this looks like a strong zone," "the market is too choppy today." Things an experienced trader evaluates in a second, but a bot needs precise numbers. Most strategies can be automated, just not "as is." They require translating intuition into measurable rules.
Key Takeaways:
- Even "discretionary" techniques can be automated, but they require precise parameter definitions, not intuition
- The real bottleneck: translating "I know it when I see it" into IF-THEN rules with specific numbers
- 5 phases from idea to live bot (2–4 months). Each catches a different category of problems. Don't skip any
- A hybrid approach (bot scans + you decide) is often the best start
Discretionary vs. Automated Trading: Why It's Not Either/Or
A discretionary trader decides based on experience. They see the chart, news, context, and in a second, they evaluate whether to enter. An automated system does the same, but needs every step written as a precise rule. Every "looks good" becomes IF condition THEN action.
| Discretionary Trader | Automated System | |
|---|---|---|
| Execution Speed | Seconds to minutes | Milliseconds |
| Emotions | Fear, greed, revenge trading | None |
| Scalability | 1 person, a few instruments | Dozens of instruments simultaneously |
| Adaptability | Reacts to unknown situations | Fails during regime changes |
| Consistency | Depends on trader's state | Identical execution every time |
| Running Costs | Trader's time | VPS, data feed, code maintenance |
I've seen stats claiming algorithms generate 60–80% of volume on major markets. In practice, that's mostly HFT and market making, not retail strategies. For a retail trader, the relevant question is different: can an algorithm execute my specific strategy better than I can?
The answer is usually not the whole strategy, but specific parts of it.
Which Strategies Can Be Automated (and How to Translate Intuition into Code)
Easily automatable strategies have clear, measurable rules: mean reversion, breakouts from defined levels, moving average crossovers, time-based strategies (e.g., opening range breakout), or grid strategies.
But even techniques that look "discretionary" can be automated if you can define them precisely. The bottleneck is almost always the same: the trader does it intuitively and can't describe it as a measurable parameter. Phase 1 of every project is exactly this translation.
| Trader Says | What We Need to Define |
|---|---|
| "I'll wait for confirmation" | Close above/below the level + 2nd candle confirms direction + volume above average |
| "This looks like a strong zone" | Swing high/low over N candles, min. distance from price in pips, swing type |
| "Market is too choppy" | ADX < 20, ATR < X% of daily average, Bollinger Band width below percentile Y |
| "I see a divergence" | Price makes new high/low, RSI/MACD doesn't. Quantify as % deviation over M candles |
| "I'll enter on a pullback" | Retrace 38.2–61.8% of last swing + rejection candle (lower wick > 60% of body) |
| "I take partial profits" | At what levels (1:1 RRR, Fib extensions, fixed pips)? What % at each level (50% at TP1, 25% at TP2, rest trailing)? Move SL to breakeven after first partial? Does trailing logic change after a partial close? |
| "The candle looks strong" | Body > 70% of total range? Body > 1.5x ATR(14)? Close in the upper/lower 25%? Volume > 1.5x 20-period average? Engulfing previous candle? Define "strong" as 2–3 measurable conditions |
| "I check the higher timeframe" | Which higher TF (M15 trader checks H1? H4? Daily?)? What specifically (trend direction via MA slope? key level proximity? RSI value?)? What when H1 says buy but H4 says sell? Define the hierarchy and conflict resolution |
Real-world example: A trader says "I enter on a pullback after a breakout." We define: breakout = close above the highest high of the last 20 candles on H1. Pullback = retrace into the 38.2–50% zone of the breakout candle. Entry trigger = first candle closing above the 50-period EMA with volume higher than the 20-period average. SL below pullback low, TP at 1:2 RRR.
5 questions before you automate:
- Can you describe your entry as an IF-THEN with specific numbers?
- Do you have fixed SL and TP rules (not "I'll move it when I feel like it")?
- Does the strategy work on a single timeframe with a single instrument? (multi-TF/pair = 3–5x complexity)
- Do you have at least 100 recorded trades for validation?
- Can you define when NOT to trade? (Filters matter more than entry signals)
A concrete example of "clear rules that are surprisingly hard to code": FTMO-style prop firm constraints. Daily drawdown is calculated from the daily starting balance (not the rolling equity peak), news blackout windows forbid open positions during high-impact releases, and some firms restrict weekend holding. Each of these is a measurable rule, but implementing them correctly requires specific logic that's easy to get wrong.
5 Phases of Automation: Don't Skip Any
Each phase exists to uncover a different type of problem. Skipping a phase = an undetected problem that costs you money.
Phase 1: Define the rules (1–2 weeks). Translate the strategy into IF-THEN statements. Deliverable: a document with precise entry/exit conditions, filters, and position sizing rules. This is where you discover how much of your strategy is actually defined and how much lives "in your head."
Phase 2: Backtest (1–2 weeks). At least 2–3 years of historical data, ideally spanning different market regimes (trending, ranging, high/low volatility). Split data into in-sample (70%) and out-of-sample (30%).
Walk-forward analysis: optimize on 6 months, test on the next 2, shift the window, repeat. Key metrics: profit factor above 1.3, max drawdown, trade count above 200, Sharpe above 0.5.
Phase 3: Paper trading (1–2 months). Demo account with real market conditions. Compare with backtest. If results degrade by more than 40%, go back to Phase 1.
Phase 4: Live with minimal risk (1–2 months). Micro lots, 0.25–0.5% of account per trade. Bot runs on a VPS with low latency to the broker (not your home PC over Wi-Fi). Minimum 50–100 live trades before scaling up.
Phase 5: Scale up (ongoing). Gradual increase at milestones, e.g., after 100 trades with equity at a new high. Volatility-adjusted sizing or fixed incremental steps.
Why not skip? A strategy that backtests great but fails in paper trading has an execution problem. A strategy that paper-trades well but loses money live has an infrastructure or psychology issue (the trader keeps overriding the bot). Each phase uncovers a different category of problems.
When Not to Automate (and the Hybrid Alternative)
Some market approaches can't be effectively automated: pure price action based on "reading the chart's story," fundamental analysis, discretionary pattern recognition without measurable parameters.
For many traders, the best start is a hybrid approach. Automation handles what it's good at, you decide where the machine falls short:
- Bot scans dozens of instruments looking for setups based on your rules
- Finds candidates → sends an alert via Telegram or email
- You assess the context (news, broader market structure, confluences)
- You approve or reject, and the bot executes with predefined SL, TP, and position sizing
- Bot manages the trade (trailing SL, partial TP, time-based exit)
Start with the hybrid approach. The bot scans dozens of instruments for you, you only evaluate the ones that pass the filter. You save hours of screening and nothing slips through.
Your bot is live. Now what? In part 2, I'll cover what goes wrong after deployment: overfitting, why backtests always lie, risk management, platform choices, and how to keep a live bot from quietly bleeding your account.
Let's discuss automating your strategy
