Risk Parity, Leverage and Trend: A Historical Test

What was tested

Seven risky sleeves: us equities, international equities, long-term us treasuries, intermediate us treasuries, reits, gold, commodity futures. The common panel contains 605 months, beginning in January 1975. Base target weights come from one equal-risk-contribution solve on the first 10 years. Those targets are then rebalanced monthly, without re-estimating their covariance. The measured period is 1985-01 to 2025-05.

The overlay averages positive trailing excess-return signals over 1, 3, 6, 12 months. Each signal uses data ending in the previous month. A sleeve receives its base weight times the fraction of positive signals; the rest stays in cash. This lag avoids using the return being predicted, but does not remove bias from choosing the rules after examining the history.

Idle cash earns the panel’s dated cash return. Borrowing above full exposure costs that return plus 0.50 percentage points a year. The benchmark starts at 42% US equities, 18% international equities and 40% intermediate Treasuries, then rebalances each January. Holdings drift between trades.

Trading costs are 10 basis points per unit bought or sold. Turnover measures actual changes from drifted holdings, including the initial purchase. Fees are debited from the cash balance at month end; any resulting debt carries financing costs in subsequent months. This is a stated monthly accounting convention, not an intraday execution model.

The sleeves are historical proxies. Gold uses a price series; these are not seven investable fund total-return records. See the long-history reconstruction for the underlying stock, bond and commodity series, and methodology for context.

The runs

Return is annualized compound growth after modelled trading and financing costs. Volatility is sample monthly return standard deviation times √12. Sharpe is mean monthly return above dated cash divided by sample standard deviation of those excess returns, times √12; undefined ratios display n/a. Worst fall uses monthly closing values, so it can miss losses within a month.

Plain risk parity’s Sharpe is 0.681 at 1× and 0.654 at 1.53×. Before spreads and trading costs, constant leverage financed at cash scales excess return and its standard deviation together. A lower ratio after costs is not evidence that diversification only works in some months.

Retrospective risk scaling

For this comparison, the leverage multiplier is the benchmark’s full-period volatility divided by each strategy’s unlevered volatility. That uses hindsight. Financing and trading costs mean the resulting volatilities are close, rather than identical; the table shows the actual values. Average exposure can be lower than the multiplier when the trend signal leaves sleeves in cash.

The scaled plain portfolio returned 0.09 percentage points a year more than the benchmark in this sample, with a -32.2% worst fall against -27.1%. The scaled overlay returned 2.32 points more, with a -15.1% worst fall. No confidence interval or withdrawal-rate conclusion is inferred from these differences.

Trading costs

At 1.53×, the overlay trades notional equal to 381% of portfolio value per year on average. These sensitivity runs keep the signal and borrowing spread fixed and vary the one-way charge. Average exposure of 93.6% does not mean borrowing is absent: costs depend on the months when exposure exceeds 100%.

These assumptions illustrate cost sensitivity. They are not measured spreads for a retail implementation, and exclude taxes, margin constraints, market impact and fund-specific fees.

Limits of the result

This is one historical path. The lookback blend and study design were selected after examining this history; the risk scaling also uses the entire measured sample. Lagged signals alone do not make it an untouched out-of-sample test. Non-overlapping decades are not automatically independent, and this page does not establish that every decade or every lookback won.

The calculation does not establish live managed-futures performance, estimate slippage from a live fund comparison or test a safe withdrawal rate. Withdrawals, inflation-adjusted spending and taxes need a separate analysis. Monthly observations also omit intramonth margin calls and drawdowns. A historical advantage here is a result to investigate, not a guarantee about a future portfolio.

Correction, 6 September 2026: actual drift now determines turnover; the benchmark genuinely rebalances annually; Sharpe uses arithmetic dated-cash excess returns. The earlier headline and claims about half-drawdowns and a half-point return gap have been replaced with the corrected figures. The measured period includes January to May 2025; complete-year statistics count only the 40 full calendar years.