Multi-Model Quantile Deep Value Assessment
Pinpoint deep value entry zones by measuring where price sits across a basket of mean-reversion models.
The framework asks: across multiple independent mean-reversion anchors, what fraction of all historical trading days has seen the asset priced lower than today? Compiling that quantile reading from four or more models—long-run moving averages, on-chain or fundamental cost-basis levels, power-law channels, analyst-derived fair-value bands—produces a composite percentile score. A score in the bottom 20% of all days signals a value zone; bottom 10% signals deep value; bottom 5% signals extreme value. Because each model uses different math and different data sources, agreement across them dramatically raises conviction. Position sizing then scales with depth: the deeper the quantile, the larger the allocation, accepting that short-term further downside is possible but long-run reversion is highly probable.
- Markets are mean-reverting: price oscillates around fundamental anchors over multi-year periods.
- A single model can mislead; a basket of independent models reduces model-specific noise.
- Quantile position is more informative than absolute price—context relative to history matters.
- Depth of value justifies position size; the rarer the reading, the larger the bet.
- Time pain and price pain are separable; cheap does not mean immediately higher.
- Probability, not certainty: the framework shifts odds, it does not predict bottoms.
- Select your model basketChoose four to seven independent mean-reversion anchors for the asset—long-run moving averages, fundamental cost-basis metrics, power-law bands, or analyst fair-value models. Prioritize anchors that use different underlying data so they are not correlated by construction.Pro tipLean toward anchors with at least a 4-year history so the quantile distribution is statistically meaningful. Mix price-derived and fundamental-derived models.WarningDo not use models that are mathematical transformations of one another (e.g., the 100-week and 200-week MA together carry less independent signal than pairing a MA with a cost-basis metric).
- Build the historical quantile distribution for each modelFor each anchor, record the asset's deviation from the anchor on every historical trading day. Rank all those deviations to produce a cumulative distribution. This tells you what fraction of all days were cheaper than today.Pro tipUse log-scale deviations if the asset compounds over time; this prevents older low-price data from distorting the distribution.
- Calculate today's percentile on each modelRead where today's price falls in each model's distribution—e.g., the bottom 12% of days for model A, the bottom 8% for model B. Record a Q-score (0–100) for each model, where lower means cheaper relative to history.WarningA model with fewer than 200 data points will produce unreliable tails; flag low-sample models and weight them less in the composite.
- Compute the composite Q-score and assign a value tierAverage the individual Q-scores (or weight by model confidence) to produce a single composite reading. Map the result: Q20 or below = value zone, Q10 or below = deep value, Q5 or below = extreme value.Pro tipIf all models agree (tight spread across the basket), conviction is higher than if models diverge by more than 15 percentile points.
- Size the position to the Q-score depthAllocate capital proportionally: a Q15 reading might justify a half-position, while a Q5 reading justifies a full or oversized allocation within your risk limits. Always retain a cash reserve because extreme-value readings can persist or deepen before reverting.Pro tipPre-define your position-sizing tiers before you are in the trade so emotion does not override the system when headlines are maximally negative.WarningDeep quantile readings do not cap downside—Q5 can become Q2. Size so that a further 20% decline does not trigger a forced exit.
- Set a reversion target and hold with convictionDefine the exit or rebalance price as the level corresponding to the median (Q50) or mean of your model basket. This is where the asset is 'fairly valued' by the same framework that identified deep value.Pro tipWrite down your thesis and Q-score on entry so you can re-read it when prices fall further and fear spikes.WarningSelling too early at Q30 because 'it recovered a bit' forfeits the bulk of the mean-reversion gain—stay in until the target tier is reached unless your fundamental thesis changes.
In early 2025, Bitcoin fell to roughly $60,000 amid a wave of fearful headlines. Analyst James Check ran his basket of mean-reversion models—realized price, true market mean, 200-week MA, power-law bands—and found the composite Q-score placed $60K in the bottom 10% of all historical days. He published a piece titled 'Welcome to Deep Value,' arguing that while further downside was possible, 90% of historical days had eventually traded higher from that zone.
A portfolio manager covers semiconductor stocks. After a 45% sector drawdown, she runs her four-model basket—price-to-book Z-score, EV/sales percentile, 200-week MA deviation, and analyst consensus fair-value gap. Three of four models show the sector in the bottom 15% of all weekly observations going back 20 years. She assigns a Q13 composite score and doubles her sector weight.
Developed by on-chain analyst James Check (Checkmate) at Glassnode and described in this BTC Sessions interview. Check applies the approach to Bitcoin market cycles, but the underlying quantile logic transfers to any mean-reverting cyclical asset.