System size
56 variables
The empirical system spans 11 Brazilian crypto ETFs and 56 variables organized across eight economically distinct predictor blocks over 1,281 B3 trading days.
Financial Econometrics · Crypto Markets
High-Dimensional Canonical Volatility Transmission in Crypto-Integrated Financial Markets
Evidence from Brazilian crypto ETFs, stablecoins, DeFi liquidity, AI-compute assets, and global macro stress.
Working Paper · Forthcoming
Full Article ↗This paper treats crypto volatility as a multi-channel transmission problem rather than a single-factor phenomenon. Brazilian crypto ETFs provide the target market, while global crypto, stablecoin liquidity, DeFi leverage, Brazilian macro fragility, global stress, safe-haven rotation, AI tokens, and AI equities form the external information blocks. The objective is to identify which latent transmission channels matter under different market states and how those channels enter both long-run and short-run volatility.
Research question
Which economically meaningful high-dimensional shock factors govern volatility in regulated Brazilian crypto ETFs, and how does that transmission change across market regimes?
Method
The HD-DCF-MTV-GARCH architecture extracts rolling Ridge-CCA canonical factors from eight predictor blocks and embeds them in a multivariate time-varying GARCH system. A Gaussian hidden Markov model identifies market regimes, while Pagan-style generated-regressor corrections and Monte Carlo experiments test whether canonical-factor significance survives first-stage estimation uncertainty.
Main findings
Dominant stress regime
11 days
The dominant non-calm regime is DeFi Deleveraging: eleven acute cascade days with an average DeFi canonical-factor z-score of -4.82, corresponding to major liquidation episodes including Terra/LUNA, Celsius, and FTX.
Generated-regressor correction
17.3 → 0.97
For BLOK11, the naive LM statistic falls from 17.3 to 0.97 after the Pagan correction, eliminating spurious significance and showing how strongly first-stage uncertainty can distort inference.
ETH persistence
0.9992 → 0.967
ETH is near-integrated under the baseline GARCH specification. Adding the canonical-factor structure lowers persistence materially, indicating that DeFi and AI-token factors absorb variation a univariate model otherwise misclassifies as persistence.
Why it matters
The paper's main contribution is attribution under dimensionality. It separates multiple economically distinct transmission channels instead of forcing crypto volatility into one narrative, while also showing that generated-factor inference can be badly misleading without correction. The result is a framework that connects high-dimensional dependence, regime classification, and volatility transmission inside one inspectable architecture.
Full research record
The external scholarly record contains the complete manuscript and citation metadata. This ShockBridge page is the research showroom summary, not a substitute for the full paper.
Open scholarly record ↗