Quantitative Research · Non-Parametric Finance
What if you could be
analytically useful
without predicting
anything at all?
Most quantitative finance demands a forecast. CPE asks a different question — and gets a more honest answer.
WHAT IS CPE? X moved big past τ days e.g. BTC ETF Y moves big next τ days e.g. Gold 97% of episodes WHAT IS LIFT? How much MORE likely given X? 0% 25% 50% 75% 100% 20% Without X (base rate) 97% Given X (CPE) 4.85×
CPE(Y, X, τ) = P( ΔYfuture > rqY | ΔXpast > rqX )
The probability Y makes a large future move, given X made a large past move — computed as a direct historical count, no assumptions.
161
instruments
Equities · Crypto · Commodities · Rates · FX · Volatility
169k
surviving signals
After CPE ≥ 0.80, lift ≥ 1.5×, n ≥ 100 observations
4.85×
peak lift
Crypto jointly conditioning gold futures over 252 days
Traditional
Fit a model. Make assumptions.
Choose a copula family
Assume stationarity
Hope it holds out-of-sample
vs
CPE Framework
Count the history. State what happened.
No distributional family assumed
Every value is a verifiable count
Honest about description vs prediction
Dr. Arun Ramanathan
zenodo.org/records/20606184 · ssrn.com/abstract=6931838
github.com/quantarram/quant-regime-research