Stationarity-Based Decision-Making: ADF Evidence and Industry-Level Portfolio Design

Abstract

This study provides a systematic portrait of the stationarity/non-stationarity status of industry indices and shows how the stationarity label governs model selection and risk control. The data comprise 36 time series at the industry/fund index level. For each series, we run the Augmented Dickey–Fuller (ADF) test with the appropriate deterministic specification (intercept or intercept trend) and AIC-optimized lags; decision rules are applied at the 1%, 5%, and 10% levels, and p-values are reported. To curb Type I error, we implement Benjamini–Hochberg FDR control alongside a battery of robustness checks (PP, KPSS, sensitivity to lag/specification, and structural-break tests). Results indicate that a decisive majority of series are stationary (30 industries at 1%, 3 at 5%, and 1 at 10%); only two categories are nonstationary. We identify a “very strong core” of eight industries with ADF statistics below −7, signaling powerful mean reversion. The methodological implication is clear: for the stationary bulk, level-based modeling with conditional volatility filters (the GARCH family) and dynamic risk metrics (VaR/ES) is recommended; borderline cases should be handled with greater operational caution; and for the nonstationary cases, trend/regime-based approaches and multivariate risk measures (e.g., CoVaR/copula-based methods) are more suitable. The paper’s contribution is to deliver a precise stationarity map at the industry level and translate it directly into actionable rules for modeling and risk management—thus bridging the gap between “unit-root testing” and “practical portfolio decisions.”