Statistical Validation
Out-of-Sample Tests
Next-generation OOS validation with calibrated ensemble, conformal prediction intervals, and multi-protocol walk-forward analysis. All computation runs in-browser.
Test Methodology Specification
NIV Engine
// Master Equation
NIV = (u × P²) / (X + F)^η
where:
u = tanh(1.0·dG + 1.0·dA - 0.7·dr)
P = (Investment × 1.15) / GDP
X = 1 - (TCU / 100)
F = 0.4·YieldPen + 0.4·max(0,RealRate) + 0.2·Vol
η = 1.5, ε = 0.001
Ensemble Architecture
Base Learner 1: L2-regularized logistic regression with class weighting (handles 7.9% base rate)
Base Learner 2: Gradient boosted decision stumps (AdaBoost, 25 rounds)
Base Learner 3: Feedforward neural network (12→8→1, ReLU, manual backprop)
Calibration: Isotonic regression (Pool Adjacent Violators)
Uncertainty: Adaptive conformal prediction (90% coverage)
12 Component Features
niv_smoothed12-month SMA of NIV
niv_rawRaw NIV score
thrustNIV thrust component
efficiency_sqEfficiency squared (P²)
slackCapacity utilisation gap
dragYield curve friction
spreadYield penalty - real rate
real_rateReal interest rate
rate_volRate volatility
niv_momentum3-month NIV change
niv_accelerationMomentum change
niv_percentileExpanding percentile rank
FRED Data Series (1970-Present)
GDP Growth:A191RL1Q225SBEA
M2 Money Supply:M2SL
Fed Funds Rate:FEDFUNDS
Investment:GPDIC1
Real GDP:GDPC1
Capacity Util:TCU
Yield Spread:T10Y3M
CPI Inflation:CPIAUCSL
Interactive Testing
Run Tests
Execute validation tests against live FRED data. All computation runs in-browser.