Why we classify regimes instead of forecasting prices

The single most expensive mistake in institutional research is treating a price forecast as a tradeable signal. Forecasts have well-known statistical pathologies: they overfit to the most recent regime, they decay rapidly, and they encourage position sizing that is structurally larger than the conviction underneath.

Regimes are different

A regime is a low-dimensional classification of the cross-asset state. It does not predict price. It conditions decisions. The same trade can be correct in one regime and incorrect in another. Identifying the regime is upstream of every reasonable position-sizing decision a desk makes.

Three buckets, recomputed every cycle

Nebula's reference framework uses three buckets: risk-on, neutral, and risk-off. The classifier runs on a small set of cross-asset inputs: realized volatility composites, cross-sectional momentum, term-structure slope, and a credit-equity divergence indicator. The framework is recomputed every cycle. There is no implied forecast horizon.

• Risk-on: cross-asset momentum positive, vol composite below its long-run median.
• Neutral: indicators inconsistent or near their thresholds.
• Risk-off: cross-asset momentum negative or vol composite above its long-run median.

What this changes operationally

When the regime is the conditioning variable, the desk's research output becomes a set of policies rather than a set of calls. Position sizing maps to regime confidence. Risk limits move with regime classification. The audit trail captures the regime in which any decision was made, which is the most useful single input to performance attribution after the fact.

Forecasts are fragile. Regimes are tradeable context.

Nebula research desk, standing position