Kernel-based identification of the perceived market value: the Bitcoin case

Modern financial markets, particularly the cryptocurrency ecosystem, are characterized by high-frequency stochastic noise and non-stationary dynamics that render traditional technical indicators structurally inadequate due to inherent phase lag. This thesis proposes a novel quantitative framework for extracting the ``Latent True Price'' of Bitcoin (BTC/USDT) by bridging the empirical mechanics of Market Profile theory with the mathematical rigor of Regularized System Identification. The core of the methodology defines the Price Weighted Mean Of Volumes (PWMOV) as the fundamental exogenous input driving structural price formation. To address the ill-posed nature of the identification problem, we utilize innovative Bayesian regularization within Reproducing Kernel Hilbert Spaces (RKHS), deploying advanced Stable Spline kernels to optimize the bias-variance trade-off. Furthermore, the severe non-Gaussianity of the order flow is rigorously addressed through a Gaussian Scale Mixture (GSM) hypothesis and an event-triggered windowing algorithm. Empirical results demonstrate that the identified optimal kernel successfully decouples macroscopic market impact from microstructural friction. To provide definitive economic validation, the reconstructed signal is first rigorously benchmarked against classical Simple Moving Average (SMA) models. By actively eliminating the phase lag inherent in traditional filters, the kernel-based framework decisively outperforms the baseline. To translate this mathematical edge into a resilient, production-grade architecture, the system is subsequently upgraded with a Dual-Layer Degradation Monitor (DLDM) (tracking Statistical and Financial Health). The final governed framework achieves an exceptional Calmar Ratio of 1.72 and constrains the maximum drawdown to 6.80% over a seven-year multi-regime horizon, empirically proving the structural superiority of trading the latent fundamental value.