Regime Classifier
A daily cross-asset reading of risk-on or risk-off: the arithmetic mean of a macro panel and an on-chain panel. A third panel, equity factor, is tracked beside them for context and left out of the composite. All three recompute every 24 hours from open data, with point-in-time inverse-correlation weights.
Daily regime classification
History
- Percentile of its last 3 years
Indicators
Backtests
Loading the backtests
No data available
Growth of $1
What a strategy holds in each regime
| StrategyEvery strategy holds the three mixes above. They differ in which reading picks the regime. The last two rows are baselines that never switch. | |||||||
|---|---|---|---|---|---|---|---|
How to read this
The out-of-sample window starts Feb 1, 2024. Risky legs earn each asset's spot return; the cash leg earns the daily-compounded DTB3 (3-month T-bill) yield. Transaction cost = ½ × Σ|Δw| × 10 bps, charged only when the bucket changes. The chart background is shaded by the composite regime: a green wash for risk-on, a warm one for risk-off, and neutral left clear.
Reading this honestly: a strategy that beats the HODL baseline on Sharpe or max drawdown but loses on CAGR is managing risk, not finding alpha. One that beats it on CAGR is either catching real signal or overfit, and the in-sample and out-of-sample split tells them apart: a big gap means overfit, a consistent result means plausible signal. None of this controls for path dependence across the single 8-year window.
Methodology
Composite = 0.5 × macro + 0.5 × on-chain. Each panel is a weighted mean of sign-aligned 3-year rolling percentile ranks; weights within a panel are inversely proportional to each indicator's mean absolute correlation with the rest of the panel. A composite percentile below reads risk-off and above reads risk-on.
Panels
| Panel | Reads | Indicators | Sources | In the composite |
|---|---|---|---|---|
| Macro 8 indicators | Upstream causes: monetary policy, credit conditions and the labor market. | 10y–2y curve, real 10y, 5y breakeven, HY OAS, DXY, initial jobless claims, VIX, copper/gold | FRED and Yahoo | Yes, half |
| On-chain 10 indicators | Crypto-native flows and network health. | DeFi TVL (level + 90d growth), stablecoin float (level + 90d growth), BTC/ETH active addresses, BTC MVRV, BTC/ETH ratio, ETH 50d/200d trend, new DEX pools per 24h | DefiLlama, blockchain.com, Coinmetrics, Yahoo, GeckoTerminal | Yes, half |
| Equity factor 8 indicators | Price-setters who have already digested the upstream inputs and are committing capital. Sign-stable signals of equity risk appetite, style and valuation. | High-beta vs low-vol (SPHB/SPLV), momentum (MTUM/SPY), growth vs value (IWF/IWD), utilities leadership (XLU/SPY), staples vs discretionary (XLP/XLY), S&P 500 50d/200d trend, IWM/SPY breadth, Shiller CAPE | Yahoo Finance and a mirror of the Shiller spreadsheet | No, tracked beside it for context; the conservative and aggressive backtests do use it |
Macro and on-chain are two independent viewpoints, averaged into the composite. The research write-up compares the composite with and without the equity factor panel, panel by panel.
How it works
- Normalization
- A rolling 3-year percentile rank per indicator, sign-aligned so +1 always means risk-on.
- Weighting
- Point-in-time inverse correlation within each panel. Each day's weights use only the trailing 3-year window, so there is no look-ahead. Redundant inputs are discounted automatically, without manual sub-grouping. No indicator carries more than 25%.
- Composite
- The arithmetic mean of the macro and on-chain panel indices (the equity factor panel is excluded), ranked as a 3-year rolling percentile and bucketed at 0.33 / 0.67.
- Smoothing
- The published label switches after 5 consecutive trading days in the new bucket, or on a single-day move in the composite percentile beyond 2σ of its trailing 1-year daily changes, in the new bucket's direction. It filters out flips at a boundary and still follows a fast, real shift.
- Storage
- The history is kept daily in long format; the current snapshot is served from the analytics store.
Read more
- The 26 indicators, in plain language: what each one is, how it is derived and how to read it.
- Regime Detection research brief: the prior-art survey and design-space mapping.
- The original two-panel classifier, macro and on-chain only, kept as a reference.