Dynamic Regime Identification and Risk-Gating in Spot Markets Using Wavelet-KAN Latent Mapping

Research Summary This paper introduces Centinel VIE333, a risk-management framework designed to solve the problem of catastrophic capital drawdowns in quantitative trading. Traditional models often fail because they attempt to trade in every market environment, even during high-risk macroeconomic shifts. Using a Wavelet Kolmogorov-Arnold Network (Wav-KAN), this architecture maps market data onto a two-dimensional "Regime Plane" defined by trend strength and volatility. Instead of just predicting price movement, the model identifies "stormy" market states and uses a probabilistic killswitch to halt trading and move the portfolio to cash during high-risk periods. Key Results Drawdown Reduction: The system successfully reduced Maximum Drawdown (MDD) in out-of-sample testing from -51.25 to -32.07. Regime Mapping: The model demonstrated a clear geometric separation between stable market states and dangerous, high-volatility environments. Capital Preservation: The findings prove that for spot markets, knowing when not to trade is more effective for long-term capital preservation than continuous forecasting.