strategy-dev

Forecasting — read probabilities + Monte Carlo without fooling yourself

How to use P(Up), Expected Change, Risk Level, and the 5/50/95 bands as decision inputs (not predictions)

7 min · intermediate

What you'll have when finished

  • Read P(Up) correctly — what 60% actually means and doesn't mean
  • Use Monte Carlo bands to set realistic TP / SL targets
  • Recognize when the model is genuinely uncertain (and act accordingly)
  • Match the right horizon (1h / 4h / 24h) to your trade duration
  • Avoid the most common probabilistic-forecast misreads

Before you start

  • No model predicts the future. Forecasts give probability distributions, not point predictions.
  • Probability is not certainty. 70% P(Up) means the move down is still 30% — and 30% is one trade in three.
  • Bands WIDEN over time. A 24h forecast is intrinsically less precise than a 1h forecast.

Walkthrough

  1. Pick the horizon FIRST — match it to your trade duration

    Forecasting has 3 horizons: **1h / 4h / 24h**. Pick the one that matches how long you actually plan to hold: - **1h** — scalp or short-term tactical trades. Most accurate (~65-70%). Smallest expected moves. - **4h** — intraday swing trades. ~60% accuracy. Moderate expected moves. - **24h** — overnight or daily-bar trades. ~55% accuracy. Largest expected moves but least precise. **Anti-pattern:** loading a 24h forecast to make a 1h trade decision. The 24h prediction averages over a full day; your hour-long window could land anywhere within that range. **Always match horizon to hold time.** If you plan to hold 6 hours, the 4h forecast is your reference (it's the closest available; switch to 24h if your hold is closer to a day).

    Success criteria: Picked horizon matching your trade duration · Did NOT just look at the longest horizon

  2. Read P(Up) honestly

    P(Up) is the probability that price closes HIGHER than now at the end of the horizon. **What 60% P(Up) actually means:** - "If we ran this exact setup 100 times, ~60 of them would close up." - Of the 60 up, some are tiny (+0.1%) and some are big (+5%). Average is around the Expected Change. - The 40 down include both small dips and bigger drops. **What 60% does NOT mean:** - It does NOT mean "price will go up by 60%." - It does NOT mean "you have a 60% chance to make money" (depends on your TP/SL). - It does NOT mean you should take the trade automatically — even strong signals fail 30-40% of the time. **Action thresholds:** - **P(Up) > 70%** — strong directional signal, trade with normal sizing - **P(Up) 60-70%** — moderate signal, half-size or wait for confluence - **P(Up) 50-60%** — barely above coin-flip, NOT a tradeable edge alone - **P(Up) < 50%** — bearish lean (the engine sees the opposite)

    Success criteria: Understood the distinction between probability and prediction · Set a personal P(Up) threshold for action

  3. Expected Change vs Expected Range — these tell different stories

    **Expected Change** = the MEAN of the probability distribution. The most likely SINGLE number outcome. **Expected Range** = the typical ±band around that mean. Like "expected ±2%." **Example:** - Expected Change: **+2%** over 24h - Expected Range: **-3% to +7%** This says: "the center of the distribution is +2%, but it's reasonable to see anywhere from -3% to +7%. Moves outside that range happen ~30% of the time." **Use case:** - Set your **TP** near the upper end of the range (e.g. +6%) so you capture upside without being greedy - Set your **SL** near the lower end (e.g. -2.5%) so you exit before catastrophic loss - Size your position so a stop-out is acceptable — DON'T size for the upper end of the range **Critical insight:** if the Expected Change is +2% but the Expected Range is ±10%, the SIGNAL is much weaker than the headline suggests. The +2% is buried in noise. Wait for a setup where range is small relative to expected change.

    Success criteria: Read Expected Change AND Range together · Set TP/SL based on the range, not the center

  4. Risk Level + Market Regime — context for the probability

    Two badges that change how you should READ the probability: **Risk Level (Low / Medium / High / Extreme):** - Low — current volatility is calm. Tighter stops are OK. - Medium — normal volatility. Use default position sizing. - High — elevated volatility. WIDEN stops or REDUCE size — same probability number means a bumpier ride. - Extreme — volatility regime change. Strongly consider sitting out. **Market Regime:** - **Trend-Up / Trend-Down** — momentum strategies work; Grid Bot doesn't - **Sideways** — Grid Bot / mean-reversion thrives; trend strategies bleed - **High-Volatility** — Algo Orders (Iceberg) to hide size; wide stops only - **Accumulation** — smart money building positions; bullish bias over weeks - **Distribution** — smart money exiting; bearish bias over weeks **Combine:** a "70% P(Up)" in a Trend-Up + Low Risk regime is very actionable. The same "70% P(Up)" in a Distribution + High Risk regime is the engine fighting headwinds — heavily discount.

    Success criteria: Read Risk Level + Regime · Adjusted your conviction based on the context

  5. The 3 timeframe cards — spot when horizons disagree

    The 3 Forecast Cards (Next 1h / Next 4h / Next 24h) show all three horizons side-by-side. **Same direction across all 3** — strong consensus. The engine sees the same story at multiple scales. Highest-conviction setups. **Short-term vs long-term disagree:** - **1h Bullish + 4h Neutral + 24h Bearish** — usually a near-term bounce inside a downtrend. The longer horizon typically wins for hold times >4h. - **1h Bearish + 4h Neutral + 24h Bullish** — short-term dip, longer-term uptrend intact. Buy-the-dip setup. **All 3 Neutral / PENDING** — genuinely no edge. Sit out. **Confidence badge on each card:** - High = engine is highly committed - Medium = leaning but not certain - Low = barely above noise - Pending = data still loading or insufficient **Action rule:** the highest-conviction trade is when ALL THREE horizons agree AND all three confidence badges are HIGH.

    Success criteria: Read all 3 horizon cards · Identified whether they agree or disagree

  6. Monte Carlo Simulation — what 1,000 paths tell you

    The chart shows 1,000 simulated price paths starting from current price, projected over the horizon. Three colored lines mark the percentile bands: - **Red dashed (5th percentile)** — only 5% of simulations went WORSE than this. Use as your defensive stop-loss reference. - **Green solid (50th, median)** — the middle outcome. Use as your realistic TP reference. - **Purple dashed (95th percentile)** — only 5% went BETTER than this. Use as your moon-shot upside. **8 faint gray lines** = sample individual paths. They give you an intuition for what individual simulations look like — most stay in the band, some break out. **How to read the SHAPE:** - **Narrow bands** — high consensus, low uncertainty. Tighter stops/TPs are appropriate. - **Wide bands** — high uncertainty. Use wider stops, smaller size, or sit out. - **Bands widening rapidly over time** — uncertainty compounds. Beyond hour 4, the 24h bands are MUCH wider than 1h. Normal. **Common misread:** treating the median as a price target. The median is the AVERAGE outcome — half the time price ends BELOW it. Don't set TP at the median expecting price to reach it; price reaches it only ~50% of the time. **Better:** set TP around the 65th-75th percentile area (between median and upper band) so you capture most realistic upside without being greedy.

    Success criteria: Identified band width (narrow/wide) · Set TP between median and 95th percentile

  7. Putting it together — the conviction checklist

    Highest-conviction trade setup checks ALL of these: ✅ P(Up) > 70% (or P(Down) > 70% for short) ✅ All 3 timeframe cards agree on direction ✅ At least 2 horizon cards show HIGH confidence ✅ Market Regime supports the direction (Trend-Up for longs, Trend-Down for shorts) ✅ Risk Level is Low or Medium (not High/Extreme) ✅ Monte Carlo bands are reasonably narrow (consistent direction) ✅ Expected Range is small relative to Expected Change (low noise) When most or all of these align, normal sizing is appropriate. When 2-3 align, half-size is appropriate. When 0-1 align, sit out. There's no edge today on this pair. **The honest truth:** strong setups are rare. The platform-wide accuracy stats (visible at [Forecast Accuracy](/forecast-accuracy)) show that even with all confluence, the engine is right ~70-75% of the time. That means 25-30% of even your best setups will lose. Size accordingly.

    Success criteria: Applied the conviction checklist · Decided act / half-size / sit out

What's next

Once your conviction checklist is satisfied: - **Single-shot trade with TP/SL** → [Smart Trade](/smart-trade). Set TP at 65-75th percentile of the MC range; SL at 5th-10th percentile. - **Want to accumulate gradually** → [Algo Orders → TWAP](/algo-orders) for execution; [DCA Safety](/dca-safety) for scheduled accumulation. - **Want to automate the regime check** → [Conditional Bot](/conditional-bots) runs similar regime classification on a timer. For platform-wide forecast accuracy (so you know whether to trust any individual call), see [Forecast Accuracy](/forecast-accuracy) — Wilson-bounded stats per timeframe.