📚 Background
Grant and Lebo (2016) offer an optimistic view of fractionally differencing short time series, while Keele, Linn, and Webb (2016) caution that estimates of fractional integration will be highly uncertain in short series. This study tests these competing recommendations by simulating fractionally integrated processes and comparing alternative modeling choices.
🔬 How the Simulations Were Set Up
- Thirty-two distinct simulation conditions were used to generate fractionally integrated time series.
- Variation across conditions included sample size and the presence or absence of short-run dynamics.
🧪 Which Models Were Compared
- General Error Correction Model (GECM), which ignores fractional integration.
- Models that apply fractional integration methods, specifically using fractionally differenced data for estimation and forecasting.
🔑 Key Findings
- Short-run effect estimates: Both modeling approaches produced similar estimates of short-run effects across conditions.
- Long-run prediction when no short-run dynamics are present: Models using fractionally differenced data yielded superior long-run predictions for all sample sizes examined.
- Long-run prediction when short-run dynamics are present: The GECM outperformed fractionally differenced models, but only for time series with fewer than 250 observations.
📌 Why It Matters
- The results reconcile the prior disagreement: fractional differencing helps recover long-run relationships reliably when short-run dynamics are absent, even in short samples; however, when short-run dynamics matter and samples are small (under 250 observations), ignoring fractional integration via a GECM can produce better long-run performance.
- Analysts working with potentially fractionally integrated short time series should weigh the expected presence of short-run dynamics and sample size when choosing between fractional integration methods and the GECM.
