Levering Risk Parity Only Works With a Trend Overlay
The case against levering a risk parity portfolio is that you raise the return and the volatility together, so the tail gets worse and nothing has been gained. Over forty years of monthly data that case holds, and the margin is thinner than its defenders suggest. It stops holding once a trend overlay is applied first, and the reason is visible in the drawdown rather than the return.
What was tested
Seven risky assets, monthly total returns, with cash as the risk-free leg: us equities, international equities, long-term us treasuries, intermediate us treasuries, reits, gold, commodity futures. The panel runs 605 months. Base weights come from a single equal-risk-contribution solve on the first 10 years, then held, which leaves 1985-01 to 2025-05 to measure. Re-solving more often was tested and made things worse, monotonically, because risk parity sizes on recent volatility and therefore loads into whatever has been calm.
The trend overlay holds each sleeve only while its trailing return over cash is positive, blended across 1, 3, 6, 12 month lookbacks, so a sleeve is held at its base weight times votes over four. Leverage applies to the whole portfolio. Borrowing above full exposure costs cash plus 0.50 percentage points and anything under full exposure earns the cash rate, so financing is charged asymmetrically rather than assumed away. Turnover is charged at 10 basis points round trip.
The runs
| 1985-01 to 2025-05 | Return | Volatility | Sharpe | Worst fall |
|---|---|---|---|---|
| 60/40annual rebalance | 8.62% | 8.72% | 0.623 | -29.2% |
| Risk parityheld, no overlay | 7.67% | 6.25% | 0.716 | -23.9% |
| Risk parity with trendoverlay read monthly | 7.45% | 3.71% | 1.148 | -4.9% |
| Risk parity, levered, no trend1.53x, no overlay | 9.65% | 9.54% | 0.677 | -35.4% |
| Risk parity with trend, levered1.53x with overlay | 9.68% | 5.64% | 1.151 | -8.0% |
Sharpe uses the panel’s own cash series as the risk-free rate. Worst fall is the deepest peak-to-trough decline in monthly closing values.
Leverage alone is a poor trade
Levering the plain risk parity portfolio makes its Sharpe ratio fall, from 0.716 to 0.677, because the financing is paid in every month while the diversification benefit is not collected in every month. Set against a 60/40 at matching risk the picture is worse than it first looks.
| All at 8.72% volatility | Exposure | Return | Worst fall |
|---|---|---|---|
| 60/40 | 100% | 8.62% | -29.2% |
| Risk parity, levered | 140% | 9.16% | -32.6% |
| Risk parity with trend, levered | 144% | 12.97% | -12.7% |
Levering the plain portfolio earns 0.53 points a year over the 60/40 and takes a deeper worst fall to get them, -32.6% against -29.2%. Anyone drawing an income from the portfolio is being asked to accept a worse sequence of returns for a return improvement inside the noise. That is a bad trade and the objection to it is correct.
The overlay changes what is being levered
Applying the trend overlay before any leverage cuts volatility from 6.25% to 3.71% and the worst fall from -23.9% to -4.9%, while giving up almost nothing in return. Levering that back to the 60/40’s risk earns 4.35 points a year and still lands a worst fall of -12.7%, less than half the 60/40’s.
The distinction worth holding onto is that leverage is not the improvement. The overlay is. Leverage only converts a ratio advantage into a return, and there has to be a ratio advantage there first. Levering a portfolio whose Sharpe ratio is 0.72 produces a levered portfolio with a slightly worse one; levering a portfolio at 1.15 holds it roughly flat, which is the signature of leverage doing its job cleanly.
What actually threatens the result
Not financing. Average exposure sits near 93% and a wider borrowing spread barely moves the answer. Trading does. The overlay turns the book over 360% a year, so the cost assumption is load-bearing:
| Round-trip cost | Return | Sharpe |
|---|---|---|
| 0 bp | 10.07% | 1.220 |
| 5 bp | 9.88% | 1.186 |
| 10 bp | 9.68% | 1.151 |
| 20 bp | 9.29% | 1.081 |
| 40 bp | 8.51% | 0.943 |
Gold and commodities are where forty basis points is plausible rather than pessimistic, and at forty the Sharpe ratio falls below one. Anyone reproducing this with retail spreads should expect to land nearer the bottom of that table than the top.
What this does not show
It is one path through one set of decades. Forty years of monthly data contains roughly four independent ten-year periods, so the honest claim is that the overlay won in every independent decade the data holds, not that it won in hundreds of overlapping windows. The blend of lookbacks was also chosen on this same data, which flatters it, though each lookback worked on its own.
It is also not a withdrawal-rate result. Everything above is compound return, volatility and drawdown. Sequence risk is path dependent in a way those summary numbers do not capture, and a shallower drawdown is suggestive rather than conclusive. Running this through a withdrawal simulation is a different exercise and would be worth doing.
Finally, the live record of trend-following is worse than backtests of it. Realised Sharpe ratios in managed futures sit nearer 0.3 to 0.5 than the figure here, and trend had a poor decade through the 2010s. Treat the gap between those numbers as the cost of implementation and slippage rather than as evidence that one of them is wrong.
Frequently asked questions
Does leverage improve a risk parity portfolio?
Not on its own. Levering an equal-risk-contribution portfolio up to the volatility of a 60/40 earned about half a percentage point a year more than the 60/40 over 1985 to 2025, and did so with a deeper worst drawdown. The extra return is close to the error bars and the tail gets worse, which is the wrong trade for anyone drawing an income from the portfolio.
What changes when you add a trend overlay?
The result inverts. Holding each sleeve only while its trailing return over cash is positive, blended across one, three, six and twelve month lookbacks, roughly halves the portfolio's volatility before any leverage is applied. Levering that back up to the same risk produced a materially higher return and less than half the drawdown. The overlay, not the leverage, is doing the work.
Is this a safe withdrawal rate result?
No. Everything here is measured in compound return, volatility, Sharpe ratio and drawdown. None of it has been run through a withdrawal simulation, and sequence risk is path dependent in a way these summary statistics do not capture. A lower drawdown is suggestive but it is not the same claim.
What would break this result?
Trading costs, first and foremost. The strategy turns over a great deal, and at forty basis points of round-trip cost the Sharpe ratio falls below one. Gold and commodities are where a spread that wide is plausible. Second, the live record of trend-following is worse than any backtest of it, with realised Sharpe ratios nearer 0.3 to 0.5, and trend had a poor decade through the 2010s.