Why Unstable Correlations Complicate Levered Stock-Bond Portfolios
Summary
The document considers whether an investor comfortable with all-equity exposure could hold a diversified stock-bond portfolio and use leverage to bring its volatility up to the level of stocks. The proposed rationale is that the mix may offer better risk-adjusted returns, provided borrowing costs remain below expected returns. The response focuses on the difficulty of estimating the correlations needed to size that leverage. Future correlations are unknown, while historical estimates have sampling error that declines only gradually as the data sample grows.
Correlation also changes across market conditions. Equity correlations can rise substantially during crises, and stock-bond relationships can shift in either direction depending on the crisis. A leverage ratio calibrated to calmer historical data may therefore miss its intended volatility target precisely when risk matters most. The answer cautions that leverage adds operational complexity as well. It does not quantify expected returns, financing costs, or crisis outcomes, so it presents a practical risk argument rather than a complete portfolio comparison.
Key ideas
- Leverage can scale a stock-bond portfolio to a target volatility, but its calibration depends on estimated covariances.
- Historical correlation estimates contain substantial sampling error at practical sample sizes.
- Correlations can change sharply during crises, undermining volatility targets based on calmer periods.
- Leveraged portfolios add operational complexity alongside model and estimation risk.
Tags
Full text
# Answer by Brian B (score 5, accepted)
# If someone is comfortable with 100% stocks, why shouldn't they just build a stock/bond portfolio and lever it to match the volatility of 100% stocks?
These are my assumptions
- A stock + bond portfolio over the long term has better risk adjusted returns than stocks alone.
- A person who is not comfortable with 100% equities would simply maintain a leverage ratio of 1 and some combination of both stocks and bonds.
- A person who is comfortable with 100% equities should still construct a stock/bond portfolio. They’d simply increase the leverage until it roughly matches the same stdev as 100% stocks instead of going 100% in stocks.
You folks probably already know all of this. But everytime I ask a financial adviser they do not recommend leverage at all.
But it seems like the reasoning is sound. As long as the cost of borrowing (LETFs, box-spreads, future etcs...) is lower than the average returns, a levered stock/bond portfolio would actually be the more rational thing to do rather than be in 100% equities.
## Answer by Brian B (score 5, accepted)
https://quant.stackexchange.com/a/78872
I think @nbbo2 nailed the answer to this question in comments (upvoted). But, to answer the question solely by its title:
One reason you would not run a levered portfolio with matched volatility is that correlations are quite unstable.
All the mathematics that one uses for running this kind of levered portfolio requires correlation/covariance, and
- The correlation you should use is the future correlation, obtainable only through time travel,
- The correlation you will use comes from some historical estimator, which converges as $O(1/\sqrt{N})$ and has uncomfortably high error sizes for practical sample counts,
- Asset correlation is actually rather unstable. In particular intra-equity correlations increase greatly in a crisis. Equity-bond correlations change a lot in crises also, with the direction depending on the type of crisis.
Thus in practice the levered portfolio has a good chance of failing to meet its purpose, while adding significant operational complexity.
I don't think any financial advisors are thinking about the disadvantages of leverage on this quantitative level, but a long conversation with a smart advisor might reveal that he/she has instincts along these lines.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.