Prediction methodology

How TheWeal produces price predictions, in full. We publish this in detail because a forecast you cannot audit is just opinion dressed up as data.

The four components

1. Momentum (weight: ~40% in near-term horizons)

Momentum is the smoothed forward drift implied by recent returns. We use a weighted blend of the 7-day and 30-day percent change:

weekly_drift = 0.7 × pct_change_7d + 0.3 × (pct_change_30d / 4.3)

The annualised equivalent (weekly_drift × 52) feeds the longer-horizon calculation. Momentum dominates short-horizon predictions (24h, 7d, 30d). Its weight decays as the forecast horizon lengthens.

2. Mean-reversion target (weight: ~60% in long-term horizons)

For longer horizons, momentum becomes less reliable. Mean-reversion anchors the long-term base case on a price level derived from the coin’s all-time high and its market-cap rank:

reversion_target = ATH × (0.7 + 0.3 × rank_quality)

Where rank_quality ranges from 0.15 (rank ~1500) to 0.99 (rank 1). Higher-quality coins (top 10) revert closer to their ATH; long-tail coins revert lower. This reflects empirical reality: top-10 coins tend to revisit prior ATHs within a multi-year cycle; sub-rank-200 coins typically do not.

3. Sentiment overlay (Fear & Greed)

The base case is then nudged up or down by the current Fear & Greed reading (alternative.me index, 0–100 scale):

sentiment_overlay = (F&G − 50) / 100

At F&G = 50 (neutral) there is no nudge. At extreme fear (F&G < 25) the base case is pulled down 5–10%; at extreme greed (> 75) it is pulled up 5–10%. The overlay’s weight scales with horizon — it has more pull on 3m–1y forecasts and less on 24h.

4. Volatility bands (bear / bull)

Around each base case we compute bear (low) and bull (high) bands using a horizon-scaled volatility multiplier:

• 24h: ±6%
• 7d: ±18%
• 30d: ±35%
• 3m: ±55%
• 6m: ±80%
• 1y: ±125%
• 2026 EOY: ±90%
• 2030 EOY: ±350%

These are heuristic — not implied volatility from options markets. Wider bands at longer horizons reflect compounding uncertainty.

Worked example: Bitcoin

Assume at the moment of publication:

• Price = $73,103
• 7d change = +2.4%
• 30d change = -4.85%
• ATH = $126,080
• Rank = 1 (rank_quality = 0.99)
• F&G = 22 (Extreme Fear)

Step 1: weekly drift = 0.7 × 0.024 + 0.3 × (-0.0485/4.3) = 0.0168 − 0.0034 = +1.34% per week

Step 2: reversion target = 126,080 × (0.7 + 0.3 × 0.99) = 126,080 × 0.997 ≈ $125,702

Step 3: 1-year base = 73,103 × 0.35 + 125,702 × 0.65 = 25,586 + 81,706 = $107,292, then × (1 + (22-50)/100 × 0.10) = $107,292 × 0.972 = ~$104,288

Step 4: 1-year bands = bear $104,288 × (1 – 1.25) = ($) cap floor; bull $104,288 × (1 + 1.25) ≈ $234,648

The actual page for Bitcoin shows these numbers — and updates them whenever the price feed refreshes. You can verify them live →.

What this model is not

It is not:

• A trading signal — it does not produce buy/sell calls
• An option pricer — it ignores implied vol, term structure, and skew
• An on-chain model — it does not incorporate flows, MVRV, NUPL, or other on-chain indicators (we may add this in v2)
• A discounted-cash-flow model — most crypto assets do not have cash flows
• Calibrated against macro variables (DXY, rates, equity vol)

It is a transparent baseline. Use it as one input among many. Cross-reference with your own analysis, the news, and — critically — your own risk tolerance.

Why we publish the formulas

Most prediction sites publish numbers without the model. Either there is no model (the number was made up), or there is a model but it is treated as trade-secret (which means you cannot audit whether it produced the number through reasoning or through wishful thinking).

We think both are bad. Predictions without auditable methodology are opinion presented as data, which is worse than opinion alone because it borrows authority it has not earned.

Historical accuracy

Starting Q3 2026, we will publish a quarterly backtest report. Every prediction made on TheWeal is recorded and grade against the realised price at the corresponding horizon. The report will show:

• Directional accuracy (% of base-case predictions that were correctly directional)
• Median absolute percentage error across all base cases
• Horizon-specific accuracy (24h vs 1y, etc.)
• Coin-specific accuracy (top 10 vs long tail)

We will not delete embarrassing predictions or selectively report. If the model is wrong, the backtest will say so.

Document control

Methodology v1.0 — published at launch. Material changes are versioned, dated, and announced in the newsletter. Prior versions are archived.