Tools
Hierarchical risk parity
Single-linkage clustering on correlation distance, then recursive bisection with inverse-variance weights; last 252 sessions; monthly. It holds 96% G-secs, 3% Nifty 500, 1% gold since May 2026. From Apr 2005 it returned 7.2% a year after tax, against 12.3% for its benchmark; its worst fall was −6%.
What it holds now
Target since May 2026: 96% G-secs, 3% Nifty 500, 1% gold.
Its record
Growth of one rupee after costs, before tax, on a log scale, beside its benchmark: equal parts Nifty 500, gold and 5-year G-secs, rebalanced each January. The benchmark holds the assets the rule chooses from but never switches, so the gap is what the rule's timing added or cost. The marks show when the rule was published and when it entered this library; only the stretch after the second mark is free of hindsight.
Hierarchical risk parity: growth of one rupee
How to read it. Log scale, so equal slopes are equal returns.
Loading chart…
Hierarchical risk parity: fall from its previous peak
How to read it. How far each was below its own previous high, weekly. The model's worst fall was −6%, against −19% for its benchmark, on daily closes; the weekly line can look a little shallower.
Loading chart…
Apr 2005 to Oct 2026. Percentages except the Sharpe ratio, turnover and tax cost. Deflated Sharpe ratio against the benchmark: 0.00 (0.95 is the usual bar), counting every rule tried.
| This model | Benchmark | |
|---|---|---|
| Return a year, before tax | 7.7 | 13.0 |
| Return a year, after tax | 7.2 | 12.3 |
| Volatility | 3.7 | 8.8 |
| Worst fall | −6.3 | −18.9 |
| Sharpe ratio | 0.3 | – |
| Average share in equity | 2.9 | – |
| Turnover a year | 0.1 | – |
| Tax cost, points a year | 0.5 | – |
The live log holds 2 sessions, 1 of them filled in after the day; its return differs from today's backtest over the same days by 0.48 points.
| Period | Dates | Total return | A year | More |
|---|---|---|---|---|
| Before the rule was published | Apr 2005 to Dec 2016 | 148.3% | 8.0% | |
| Published, before it entered this library | Dec 2016 to Oct 2026 | 98.8% | 7.3% | |
| Since it entered this library (live) | Oct 2026 to Oct 2026 | 0.1% | – |
What it held
The share of the model in each asset at every month-end.
Hierarchical risk parity: holdings at each month-end
Loading chart…
The rule
Single-linkage clustering on correlation distance, then recursive bisection with inverse-variance weights; last 252 sessions; monthly.
Lopez de Prado (2016). Rule published 2016; entered this library 2 Oct 2026. It is the portfolio lab's rule, unchanged.
The workings
Data
Every chart has a table view, a CSV download and a link to its published JSON. Bundles are listed in data status.