Macro-Financial Signals and On Chain Fundamentals in Bitcoin Predictability : Assessing the Incremental Value of U.S. Indicators for Weekly Returns (2013–2025)

Details

Supervisors

Vande Kerckhove, Corentin ; Petitjean, Mikael

Faculty

Louvain School of Management

Degree label

Master [120] : ingénieur de gestion, à finalité spécialisée

Abstract

enBitcoin’s institutionalization has turned its exposure to U.S. macro-financial conditions into a portfolio question. A substantial literature links monetary policy, inflation, the dollar, and equity factors to Bitcoin returns, but establishes that link in sample, with the benefit of hindsight. This thesis asks a narrower question: does the macro-financial block improve real-time forecasts of weekly Bitcoin log-returns beyond crypto-native variables, and does the answer vary across market regimes? Four nested specifications (baseline, crypto-native, macro-financial, combined) are estimated identically under a linear ARIMAX and a gradient-boosted tree model on 678 weekly observations spanning 2013 to 2025. A 156-week rolling window, whose structure is frozen before any access to results, yields 510 out-of-sample forecasts per configuration, compared through small-sample-corrected Diebold-Mariano tests and a Model Confidence Set, and evaluated over four event-based regimes fixed before any access to results. The answer is negative in all four. Adding macro-financial information never improves accuracy: it significantly degrades the linear forecast (DM = −2.52, p = 0.012) and leaves the tree forecast statistically unchanged (p = 0.359). The Model Confidence Set eliminates exactly one configuration, the linear combined model. No specification outperforms a rolling historical mean. In-sample detection of a macro-crypto link does not survive the cost of estimating it out of sample.