The benchmark comparison block on the Performance page prints seven figures, and two of them together say more about a book than the other five combined. On the capture I am working from, up capture reads -0.510 and down capture reads 1.214, both against SPY.
Read them as a pair. Down capture above one means the book fell further than the benchmark across the periods the benchmark fell. Up capture below zero means the book did not merely lag when the benchmark rose, it lost money while the benchmark rose. There is no configuration of that pair that flatters the process. A manager who is simply defensive prints low up capture and low down capture. A manager who is aggressive prints high on both. This pair is the asymmetry running the wrong way on both sides at once.
Why the pair is a sharper diagnostic than the alpha figure beside it
The same block prints an alpha of -203.67% annualised, a tracking error of 44.47% and an information ratio of -6.940. Those are all conclusions. They tell an investment committee that the book underperformed, which anyone could see from the return, and they tell nobody what to change.
The capture pair is a decomposition rather than a conclusion. It splits the sample by the sign of the benchmark move and reports what the book did inside each half separately. That is the only one of these seven numbers that generates a follow-up question with an answer attached, which is where the exposure went in each half and whether the difference was intentional.
The other reason to lead with capture in a review is that it is legible without a quant present. Losing money in the up half and losing more than the market in the down half is a sentence a trustee understands. An information ratio of -6.940 is a sentence that invites a discussion about tracking error definitions.
The trade behaviour that produces this pair
Four mechanisms produce a negative up capture with a down capture above one, and they need different remedies, so the diagnosis is not complete until you know which one you have.
- Directional positioning that is inverted relative to the benchmark. Short or flat when the benchmark rallies, long when it falls. On this account the direction split reads 73 long trades at a 28.8% win rate for -44.64% and 14 short trades at a 7.1% win rate for -1.18%, so the short book is small and the losses are overwhelmingly in the long book, which points away from a simple net-short explanation.
- Exposure that scales up into weakness. Averaging down, adding on drawdown, or a position sizing rule that keeps notional constant as equity falls. This produces exactly the asymmetry in the pair, because the book carries more risk in the down half than it did in the up half.
- Forced or mechanical exits clustering in down periods. Liquidations, margin reductions and stop cascades all deliver realised losses that coincide with benchmark weakness without any intent behind the timing.
- An asset base that has nothing to do with the benchmark. On this account every one of the 87 trades sits in the crypto row of the asset class table. Measured against SPY, the capture pair is then describing coincidence rather than dependence, and the correlation figure of 0.077 in the same block says as much.
The distinction between the second and third mechanisms matters for what you write in the risk policy. Scaling into weakness is a sizing rule you can change. Forced exits in down periods are a leverage and liquidity problem, and changing the sizing rule will not touch it.
Why capture beats beta on a book with 96 days of history
The same block prints a beta of 0.187 alongside a correlation of 0.077, and the panel footer records 86 observations across a 96-day span. A beta is a regression slope, and its precision depends on shared variance. With a correlation that low, almost none of the book's variance is explained by the benchmark, so the slope estimate is resting on very little and will not be stable out of sample. Quoting 0.187 as a market sensitivity in a review is quoting a number whose confidence interval nobody in the room has computed.
Capture ratios avoid two of the assumptions that beta makes. They do not assume the relationship is linear, and they do not assume it is symmetric in the two directions. They are conditional averages, which is a much weaker claim about the world, and weaker claims survive short samples better.
They are not free of the sample problem. With 86 observations across roughly four months, the number of benchmark-down periods is small, so the denominator of the down capture is thin and one bad session can move it. Treat the pair as a diagnosis that tells you where to look rather than an estimate you extrapolate. Recompute it with the single worst day removed. If the sign flips, you have an anecdote about one day, and the correct action is to go and look at that day.
The benchmark selector is a claim, and this one is contestable
The panel offers None, S&P 500, Nasdaq 100, Bitcoin and Total US Market. The figures above were computed against SPY on a book whose asset class table shows crypto and nothing else.
That is a defensible choice only if the mandate is written against equities, in which case underperforming a passive equity allocation is exactly the thing the committee wants measured. If the mandate is written against crypto beta, SPY is the wrong yardstick and the capture pair is measuring the overlap of two processes that have no structural reason to move together.
The cheap and correct response is to run the pair against more than one benchmark and report both. If up capture and down capture change materially between SPY and Bitcoin, that difference is itself the finding, and it belongs in the review rather than in a footnote. A book that shows a bad capture pair against every available benchmark has a risk problem. A book that only shows it against one has a benchmark problem, which is a different meeting.
What to bring to the review once you have the diagnosis
Three artefacts turn this pair from an observation into an agenda item.
First, the exposure profile split by the sign of the benchmark move. Average gross and net exposure in benchmark-up periods against benchmark-down periods. If gross is higher in the down half, the asymmetry is a sizing decision and you can name the rule that produced it.
Second, an attribution of the down-half losses by venue and by direction. On this account the venue table reads Hyperliquid at 70 trades for -37.79%, Alpaca at 16 for -8.50% and Manual at 1 for +0.47%, which is enough concentration that the down-capture question is largely a question about one venue and the execution conditions on it.
Third, the counterfactual. Recompute the capture pair with a hard cap on exposure additions during drawdown, applied to the historical trade record. That is not a backtest of a new strategy and should not be presented as one. It is a bound on how much of the asymmetry the proposed control could have touched, and it is the number a committee will ask for immediately after they see the pair.
The thing not to do is to argue the ratio down. On 86 observations you can always find a reason a figure is unreliable, and here the reasons are real. But an unreliable estimate of a pair this lopsided is still a reason to open the exposure profile, and the exposure profile does not depend on the estimate at all.