Reducing Risk-Parity Concentration in Heterogeneous Portfolios
Summary
The document considers a heterogeneous portfolio in which inverse-volatility risk parity can assign very large weights to instruments whose estimated volatility has fallen relative to the rest. It describes a proposed cap that compares risk-parity weights with equal weights, but notes that a fixed cap may behave poorly as the number of instruments changes. The discussion focuses on ways to control concentration while allocating risk across unlike assets.
Suggested approaches include grouping instruments into categories and applying risk parity across those groups, and replacing variance with a downside-oriented measure such as historical value at risk, expected shortfall, or semivariance. A further response raises nonlinear volatility models, including GJR-GARCH and threshold GARCH, as ways to respond to asymmetric or rapid volatility changes. These are suggestions rather than tested comparisons: the document provides no performance evidence, calibration guidance, or rule for choosing categories, horizons, and risk limits. Any implementation would need to account for estimation error and the portfolio’s objectives.
Key ideas
- Inverse-volatility risk parity can concentrate capital in assets with unusually low estimated volatility.
- A fixed weight cap may not adapt well when the number of instruments changes.
- Allocating risk parity across asset categories is one proposed way to limit concentration.
- Downside measures such as value at risk, expected shortfall, and semivariance can replace variance.
- Nonlinear GARCH variants are suggested for capturing asymmetric or quickly changing volatility.
Tags
Full text
# Controlling portfolio concentration # Controlling portfolio concentration I'm working with a heterogenous basket of instruments (in volatility terms). Risk parity allocation seems to be useful for the portfolio( * 1/Volatility). However, there are times when the volatility of a couple of instruments drops down significantly compared to the rest of the universe. Risk parity asset allocation shifts too much capital to the low-vol instruments(20-50%) which is highly undesirable. Basic improvement to the allocation approach introducing a cutoff rule based on comparing the risk parity allocation to a naive equal weight approach and limiting the allocation to a maximum percentage(10-20%). However, this doesn't seem to be optimal especially when universe size becomes too small or too large. Any ideas/improvements/suggestions on how to deal with risk allocation in heterogenous universes while actively avoiding concentrating the portfolio weights to a few instruments? ## Answer by not.so.quanty (score 5) https://quant.stackexchange.com/a/10597 I would create categories, and work on risk parity among the categories. Otherwise, variance is not really a good measure of downside risk: Change your risk measure, use a rolling window historical VaR or Expected Shortfall at some horizon that matches your investment style. downside semi-variance could do the trick too if don't want to change your algo too much. ## Answer by uday (score 1) https://quant.stackexchange.com/a/10714 Are you working with futures data (mixing rate futures with equity futures) OR allocating between macro instruments? If so, using non-linear variants of GARCH (GJR-GARCH, TGARCH etc.) are common way to solve your risk parity allocation issue that you might be facing. One common related issue is not that the volatility to a couple of instruments drop down significantly per se as much as that the volatility of those instruments which saw a drop in volatility happened to pick up very rapid and the portfolio was unable to response to fast increase in volatility of such instruments. The non-linear variants of GARCH can resolve this too as they can pick up very fast increases or decreases in volatility than regular GARCH or EWMA volatility.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.