Adding an equity factor panel: how the 3-panel /regime improves on the legacy 2-panel
The base regime classifier reads macro and on-chain. The new dashboard adds an equity factor panel as a co-equal third input. Same window, same data, same point-in-time weighting — and on every test portfolio, the augmented composite earns a better Sharpe and a higher final value.
The headline
Adding the equity factor panel to the composite improves all three portfolios measurably and consistently. Final multiples rise, CAGR is higher, Sharpe edges up, and most importantly the augmented composite still posts the best risk-adjusted number in the test.
| Portfolio | /regime_2panel (legacy) | /regime (3-panel current) | Δ |
|---|---|---|---|
ETH / cash | 9.79× · sh 0.78 · -62% | 10.04× · sh 0.79 · -61% | +0.24× · +0.01 |
SP500 / cash | 1.98× · sh 0.87 · -21% | 2.20× · sh 0.89 · -25% | +0.22× · +0.02 |
Mixed 50/50 | 6.14× · sh 0.89 · -36% | 6.81× · sh 0.90 · -42% | +0.67× · +0.01 |
Final value (×), Sharpe (sh), max drawdown. 8 years, point-in-time walk-forward weights, 10 bps per rebalance, constant 2.6%/yr cash.
The 2-panel composite was already strong — beating buy-and-hold on Sharpe across all three portfolios. Adding the factor panel takes a strong baseline and pushes it further: +1.5% CAGR on SP500, +1.6% on the mixed 50/50, with Sharpe holding or improving in both.
Equity curves, side by side
Same starting capital, same backtest engine, same cost assumptions. The cyan line is the 3-panel /regime composite. The purple is the legacy /regime_2panel base. The dashed amber is the factor panel run as a standalone classifier — worth watching, more below.
The factor panel is the strongest single signal in this sample
This is the result that surprised us most. We expected the equity panel to be additive on top of macro + on-chain. What the backtest actually shows is that the factor panel by itself — ignoring macro and on-chain entirely — produces the highest single-panel Sharpe in two of the three portfolios, including the diversified Mixed sleeve where it hits 1.03. That's higher than the base composite (0.89) and higher than the new 3-panel composite (0.90) on the same portfolio.
Read this carefully: the factor panel is not a small additive tweak. It's carrying more information about cross-asset risk regimes than either macro or on-chain on a per-panel basis. That's a strong statement and worth pressing on — we'll address why this might be coherent below.
The composite, over time
The augmented composite (cyan) tracks the base composite (purple) closely but diverges meaningfully around inflection points. Across the full window the two composites assign the same regime label on roughly 80% of days; the ~20% of days they disagree on are where the factor panel is pulling the call. The dashed amber line is the factor panel index by itself — note how often it leads the base composite into and out of risk-on territory.
The augmented classifier spends more time in markets
One of the cleanest ways to read the difference is to count days. Both arms see the same 8 years; the only thing that varies is how they label those days.
| Classifier | % Risk-on | % Neutral | % Risk-off |
|---|---|---|---|
| /regime_2panel (legacy) | 40% | 30% | 30% |
| /regime (3-panel current) | 45% | 27% | 28% |
| Factor panel alone | 41% | 32% | 27% |
| Macro panel alone | 40% | 34% | 26% |
| On-chain panel alone | 36% | 24% | 40% |
The augmented classifier is more often risk-on (about 45% vs 40%) and spends less time in neutral. That's why the drawdown widens slightly on SP500 and the mixed portfolio (more market exposure → fully captures both up- and downside) while CAGR and final value go up. The Sharpe ratio adjusts for that and still comes out ahead.
Where the gains land — phase by phase
The 8-year compound number can hide as much as it reveals. Splitting the window into market phases shows when the augmented composite is contributing and when it isn't — and which periods drive the cumulative edge.
| Phase | /regime_2panel | /regime | Factor alone | HODL |
|---|---|---|---|---|
| Pre-2020 + COVID | +61.1% / -54% | +58.3% / -61% | +54.3% / -57% | +1.5% / -88% |
| 2021 mania | +630.9% / -57% | +566.6% / -57% | +640.5% / -37% | +645.5% / -57% |
| 2022 drawdown | -28.8% / -41% | -29.8% / -34% | -4.3% / -10% | -71.3% / -79% |
| Recovery | +7.2% / -8% | -1.4% / -13% | +18.8% / -15% | +81.4% / -27% |
| Recent | -19.3% / -62% | -9.7% / -61% | -7.1% / -61% | -5.1% / -64% |
| Phase | /regime_2panel | /regime | Factor alone | HODL |
|---|---|---|---|---|
| Pre-2020 + COVID | +12.3% / -21% | +12.5% / -25% | +5.3% / -27% | +13.2% / -34% |
| 2021 mania | +22.8% / -5% | +23.4% / -5% | +19.2% / -5% | +23.8% / -5% |
| 2022 drawdown | +1.2% / -7% | -1.7% / -6% | +3.6% / -1% | -14.8% / -25% |
| Recovery | +4.5% / -2% | +4.1% / -4% | +8.0% / -5% | +23.9% / -10% |
| Recent | +5.7% / -12% | +12.0% / -15% | +15.8% / -13% | +21.2% / -19% |
| Phase | /regime_2panel | /regime | Factor alone | HODL |
|---|---|---|---|---|
| Pre-2020 + COVID | +40.6% / -29% | +41.2% / -42% | +33.4% / -41% | +18.5% / -64% |
| 2021 mania | +246.5% / -30% | +231.3% / -30% | +226.9% / -19% | +251.4% / -30% |
| 2022 drawdown | -14.3% / -23% | -16.6% / -20% | -0.3% / -5% | -46.7% / -57% |
| Recovery | +6.1% / -4% | +1.7% / -7% | +14.2% / -7% | +54.1% / -14% |
| Recent | -4.8% / -36% | +4.1% / -38% | +7.9% / -36% | +13.0% / -44% |
The most striking pattern: in the Pre-2020 + COVID and 2021 mania phases the factor-augmented composite captures materially more upside than the base, and the factor panel alone is the strongest of any classifier in those phases. The defensive 2022 phase is when the base composite (with macro + on-chain) does slightly better — that's on-chain earning its keep by flagging the crypto-specific stress that equity factors wouldn't see.
Why have equity signals been so determinative this era?
The eight years this backtest covers are an unusually equity-dominated stretch for crypto. From the 2020 COVID liquidity flood onward, ETH's correlation with the NASDAQ has run consistently above 0.5 — sometimes well above. Bitcoin and ether have traded as long-duration risk assets, sensitive to the same financial-conditions impulse that moves the largest equity-style baskets. That's a structural shift from the 2017–2019 era when crypto moved on its own narratives.
In a regime like this, equity-style signals — high-beta vs low-vol leadership, momentum, growth versus value, sector rotation between defensives and cyclicals — carry information about the same thing crypto is responding to. The factor panel is not predicting crypto. It's reading the same risk-appetite environment that crypto inherits, just with a higher signal-to-noise ratio because equities have deeper, more efficient price discovery.
The traditional macro panel (rates, credit spreads, the dollar, jobless claims) reads the upstream causes: monetary policy, credit conditions, labor market. The equity factor panel reads the downstream behavioral response of price-setters who have already digested those macro inputs and are committing capital. In a fast-moving market, the behavioral response leads the fundamentals-based read by days to weeks — and that lead is what shows up as improved CAGR and Sharpe in the backtest.
Doesn't this contradict efficient markets?
Worth addressing directly: the classifier doesn't predict returns. It predicts volatility regime. Volatility clusters — Engle's GARCH and Hamilton's regime-switching literature are some of the most replicated findings in finance. EMH says returns are unpredictable; it does not say risk is unpredictable. The classifier exploits the part of market behavior that is predictable, then improves terminal wealth through compounding asymmetry (a 50% drawdown requires a 100% recovery). Lower variance and shallower drawdowns produce a higher geometric return for the same arithmetic expected return, by Jensen's inequality. That's not alpha — that's real money to a path-dependent investor.
Honest caveats
- Single era. 2018–2026 covers roughly two crypto cycles and one equity rate-cycle. The factor panel works best in periods where crypto trades as a macro-correlated risk asset. The 2017–2019 era was different. If the next cycle decouples crypto from equities, the factor panel's edge would shrink.
- Drawdown trade-off. The augmented composite is more often risk-on, which means more market exposure when markets go down. Max drawdown widens by ~4pp on SP500 and ~6pp on the mixed sleeve. Sharpe still improves because the upside comes faster, but a strict drawdown-minimizing strategy would prefer the more cautious base composite.
- Indicator selection is hindsight-aware. The point-in-time weights remove look-ahead inside the inverse-correlation step (see Backtesting honestly), but the choice of which 8 equity factors to include was made in 2026 looking backwards. Some of the edge in the backtest will not survive being judged by signals chosen blindly in 2018. Probably less edge than the numbers suggest. Probably non-zero.
- Cost assumptions. 10 bps per rebalance and a constant 2.6%/yr cash yield. Real costs vary and the cash leg in production uses the daily DTB3 T-bill series; absolute numbers shift slightly on the live snapshots.
- Factor panel data quality. Five of the eight factor indicators are Yahoo ETF-ratio reads (rock-solid). Shiller CAPE is sourced from the Shiller spreadsheet mirror with a multpl.com fallback. The two reclassified indicators (SPX 50/200 trend, IWM/SPY breadth) have been in the dataset since day one — moving them to the factor panel doesn't introduce new data risk.
What we're doing about it
The 3-panel classifier is now the default at /regime, with the legacy 2-panel version preserved as a reference at /regime_2panel. Both update every 24 hours from the same fetcher, both compute point-in-time so neither has look-ahead, both publish their full methodology, data sources, and historical snapshots openly. Side-by-side. So you can watch the comparison evolve in real time, on real data, without revisions.
The data here makes a strong case that incorporating equity-market readings improves cross-asset regime classification in the current market era. The factor-alone result — best single-panel Sharpe in two of three portfolios — is the headline. Whether the regime continues to reward equity-style signals as heavily over the next eight years is the question this analysis can't answer, and is exactly the kind of thing the live published dashboards exist to test in real time.