Integrating Crypto-Market Proxies as Financial Stability Drivers: A Random Forest Approach

This paper proposes a methodology of whether integrating cryptocurrency native proxies along with traditional macroeconomic indicators enhances the explanatory accuracy of financial stability modeling. The center objective of the study is evaluating the systemic risk impact of various crypto-market proxies on a constructed Financial Stability Index (FSI), by using a machine learning framework. We used a Random Forest algorithm to quantify economic stress across three distinct monetary policy regimes during the 2018-2024 window. Using this non-linear approach, we make sure that the proposed methodology captures complex transmission channels that linear econometric models might miss, allowing for a more efficient stress testing method of the cryptocurrencies ecosystem. We measure the in-sample explanatory power of financial stress using three model specifications, a Macro Baseline model (restricted to including only traditional financial indicators), a Full High-Dimensional crypto-integrated model (with the entire set of variables proposed), and an Integrated Parsimonious model consisting of the most relevant macro and crypto proxies which were derived through 5-fold cross-validated Recursive Feature Elimination (RFECV). The results indicate that the Parsimonious model outperforms the others, explaining 79.43% of FSI variance compared to 62.97% for the Macro Baseline and 75.64% for the High-Dimensional model. The feature importance analysis for the Parsimonious model identifies the 10-Year Treasury Yield and BTC-S&P 500 correlation as the main non-linear amplifiers of financial instability, while the rest of the exogenous variables act as lower, but still significant, bidirectional coincident indicators. This research points to the fact that crypto-market proxies, specifically Bitcoin volatility and stablecoin depegging events are not merely speculative noise, but distinct, non-linear amplifiers of systemic risk, validating therefore the necessity of including alternative indicators for modern financial surveillance.