Risk Everywhere: Modeling and Managing Volatility¶
Bibliographic record. Follow the original-source link for the publication.
| Field | Value |
|---|---|
| Primary domain | Volatility |
| Other domains | — |
| Methods | Research Methods |
| Facets | — |
| Authors | Tim Bollerslev, Benjamin Hood, John Huss, Lasse Heje Pedersen |
| Published | 2018-05-22 |
| Source | The Review of Financial Studies |
| Identifiers | doi:10.1093/rfs/hhy041 |
| URL | Open original source |
Editorial synthesis¶
Why it matters¶
The study turns cross-asset volatility commonality into a practical forecast-shrinkage idea: pooling centered realized-volatility models across assets can reduce asset-level estimation noise, while the smooth HExp and HExpGl forecasts connect statistical accuracy, turnover, and risk-targeting utility in one framework. (full_text:S28, full_text:S39, full_text:S48, full_text:S57, full_text:S63, full_text:S345)
Main author claims¶
- The authors construct daily risk measures from five-minute intraday realized volatility plus overnight squared returns for 58 assets: 20 commodities, 21 equity indices, eight fixed-income futures, and nine currencies. Start dates vary by asset and all series end on September 30, 2014. (
full_text:S28,full_text:S72,full_text:S78,full_text:S95,full_text:S452) - The authors report an all-assets out-of-sample R-squared of 49.2% for mega HExp and 50.6% for HExpGl with its time-zone-aware global factor in 20-day expanding-window forecasts. They attribute the gain from pooling mainly to lower forecast variance. (
full_text:S156,full_text:S287,full_text:S288,full_text:S289,full_text:S332,full_text:S333,full_text:S338,full_text:S345,full_text:S347) - The authors report about 48 basis points per year of realized-utility gain for HExpGl over a static model under an annual Sharpe ratio of 0.4, risk aversion of two, and a 20% volatility target. Relative to 21-day RV, the incremental gains are about six basis points for HExp and another two for HExpGl, with implementation examined under full-spread and gradual-trading scenarios. (
full_text:S361,full_text:S374,full_text:S377,full_text:S393,full_text:S404,full_text:S425,full_text:S434)
Data, method, or discussion scope¶
The study compares static, daily, 21-day RV, HAR, MIDAS, HExp, and HExpGl models on one historical asset panel using 20-day variance forecasts, R-squared and DM loss comparisons, and mean-variance risk-targeting utility. It evaluates volatility forecasts and exposure scaling rather than directional return alpha; no independent replication was performed here. (full_text:S287, full_text:S288, full_text:S298, full_text:S332, full_text:S334, full_text:S353, full_text:S354, full_text:S369)
Main limitations¶
Normalized realized volatilities are similar but not identically distributed, as pairwise KS tests still reject equality. MIDAS tuning uses the full sample, which the authors explicitly acknowledge is not truly out of sample. Utility depends on a constant Sharpe ratio, risk aversion, and target-volatility assumptions. Trading costs are linear and approximated by either the full spread or moving 15% toward the target each day, without nonlinear impact, capacity, or optimal execution. The supplied PDF also refers to a separate Online Appendix that was not included. (full_text:S15, full_text:S138, full_text:S269, full_text:S294, full_text:S336, full_text:S353, full_text:S356, full_text:S361, full_text:S410, full_text:S425, full_text:S433, full_text:S434, full_text:S437, full_text:S438)
Relationships¶
- None recorded.