What Equity Prop Desk Traders Should Actually Track (It Isn't Win Rate)

A note on scope before the metrics. A great many desks already compute all of this continuously, in a risk system built for exactly this purpose, and if yours does then a second tool is duplication rather than insight — a point worth taking seriously enough that it gets its own section near the end. This piece is written for the seats where that is not true: smaller desks, remote and funded equity seats, individual books inside a firm whose system reports position and risk but never decision quality, and traders who want a record that survives a change of employer.

1. What Changes When the Capital Is Not Yours

Retail metrics answer a personal question: is this working, should I keep doing it. Desk metrics answer an allocation question: what should this trader be given next quarter, and what is the risk of giving it. Three consequences follow, and they reorder the whole scorecard.

Net P&L stops being the headline. It is the output of a process, and on its own it cannot distinguish a repeatable edge from one good sector call. Capital becomes a denominator rather than a background fact, because the firm is allocating a scarce resource and cares what you return per unit of it. And positions stop being independent: a book is not a list of trades, it is a set of exposures, several of which are usually the same exposure wearing different tickers.

Win rate survives none of this. It says nothing about size, nothing about capital consumed, and nothing about whether your winners and your losers were the same bet. It is the first number most journals show and close to the last one a desk would ask for.

2. Attribution: Which Decisions Actually Produced the Month

Take a month that closes at +$4,100. Reported as one figure it is a modest, unremarkable, perfectly acceptable month. Decomposed by name it is something else entirely:

The four transport names produced +$12,600. Everything else lost −$8,500. This was not a +$4,100 month; it was one correct sector call carrying a book that was otherwise bleeding, and the net figure is the only presentation of it that hides that.

Three attribution axes are worth carrying, and only the first is standard:

• By name — the base layer. Concentration of P&L into very few names is the single most common finding, and it is the same outlier-concentration test that decides whether an edge is real or is three lucky trades.
• By sector — the equities-specific axis with no forex equivalent. This is what turns seven trades into two decisions.
• By catalyst — earnings, guidance, M&A, index events, macro prints, pure technicals. The axis nobody logs and the one that most reliably explains why a strategy stops working, because catalysts come in regimes.

A net P&L figure is a summary of a book you have not read. Attribution is reading it.

3. Return on Capital Deployed

Two traders on the same desk, same month, same P&L:

Identical on the P&L report, more than three times apart on the number that decides who should get more buying power. If the desk has a fixed pool to allocate, B is where the next dollar goes, and A's result depends on continuing to receive an allocation that is producing 2%.

Two implementation details matter. Average deployed capital must be time-weighted, not taken from the peak or from month-end — a position held two days does not consume what one held three weeks does, and using a snapshot rewards whoever happened to be flat on the measurement date. And it should be measured against capital actually tied up rather than the allocation on paper, because the gap between the two is itself the finding. The habit this metric disciplines is specific: parking size in low-conviction positions that neither lose money nor earn their keep.

4. Concentration: When Five Positions Are One Bet

Return to the month in section 2. UNP, NSC and CSX are railroads. ODFL is trucking. Four positions, each sized independently, each risking say 1% of the book — and a trader who believes they are carrying 1% per idea is in fact carrying roughly 4% on a single question about freight volumes. It worked, which is why nobody looked. The month it does not work, three or four stops resolve on the same morning.

This is the measurement gap that separates a desk book from a retail one, and it is invisible to every position-count-based risk rule. “No more than six open positions” and “no more than 2% per name” are both satisfied by a book that is one bet. What is needed instead is exposure grouped into correlated clusters, with the risk summed inside each cluster rather than across the position list.

Long/short books get the same problem in a subtler form. Long four industrials against short two staples is not market-neutral and is not two positions — it is a single cyclical-versus-defensive factor bet with a beta that is nowhere in the position list. Netting gross exposure to something small does not make a book neutral; it makes the factor concentration harder to see.

The practical version of this is a pre-trade check rather than a monthly report. The useful question is not “was I concentrated last month,” it is “does the order I am about to send join a cluster that is already heavy,” asked at the moment the size is still editable.

5. Borrow and Locate as Per-Trade Fields

Short financing is the cost most likely to be missing from a trader's own record, because it does not arrive per trade. It arrives as a monthly financing line that nobody attributes back to the positions that caused it, which means the strategy's measured expectancy is gross and its real expectancy is something lower that nobody has calculated.

The arithmetic is unforgiving on tight names. A $100,000 short held 21 days at a 40% annualized borrow rate costs 100,000 × 0.40 × 21/365 = $2,301. If the strategy's average winning short makes $3,000 gross, borrow has taken roughly three quarters of the edge, and the journal that records only entry and exit price will never show it. Rates on hard-to-borrow names range from a few basis points to triple digits annualized, move daily, and differ between prime brokers — so treat any single figure as an example rather than a benchmark and take the real number from your own locate desk.

Three fields close the gap: the borrow rate at entry, whether the locate was general collateral or hard-to-borrow, and the accrued financing at exit. With those, short expectancy can be computed net, and the common finding — that the hard-to-borrow subset has materially worse net expectancy than the general-collateral subset despite better gross numbers — becomes visible instead of theoretical.

6. Execution Quality, and Which Benchmarks You Can Actually Compute

On a real desk, execution is measured against a market benchmark: arrival price (the mid at the moment the decision was made) or interval VWAP (the volume-weighted average over the working period). Both require a market data feed alongside your fills, which is exactly why they live in the firm's execution stack rather than in any journal.

It is worth being precise about the distinction, because the two answer different questions. Benchmarked slippage asks whether your execution was good relative to what the market offered. Trigger-based slippage — the difference between your fill and the price your own order was supposed to trigger at — asks whether your plan survived contact. The second is computable from your own records with no market data at all, and for a discretionary trader it is arguably the more actionable of the two: it catches stops that fill three cents through on thin names, limit orders that never get the price you assumed when you sized the position, and the systematic gap between the trade you designed and the trade you got.

MAE and MFE sit alongside it and need no feed either. On intraday equity books the highest-value cut is MAE by liquidity band: if adverse excursion on your thinner names is consistently deeper before the eventual winners resolve, the stop is being placed against a spread rather than against an invalidation level, and the fix is sizing rather than a tighter stop.

7. When Your Firm's System Already Does This

Then use it. A desk risk stack that computes attribution, exposure and financing continuously is better at this than any journal will be, because it sits on the fills and the market data and updates in real time. One equity trader we spoke to described his firm as running “its own trading terminal and its own accounting and metrics — closed ecosystem,” and for that seat the honest answer is that a second tool adds nothing.

The test is concrete. Can you answer, from screens you already have: what was my return on deployed capital last month, and which correlated cluster produced my largest drawdown? If yes, stop here. If the system reports position and risk but never decision quality — what you expected of a trade, why you took it, what the catalyst was, whether you followed the plan — that gap is real, and it is the one an individual record fills. The other two cases are portability, since a track record living in an employer's database does not travel when you do, and traders running a personal book alongside a desk seat who need both measured the same way.

What to Log, Specifically

• Sector and catalyst per trade — the two attribution axes that cannot be reconstructed later from a fill file.
• Capital deployed and holding period — together these give time-weighted average deployed capital, which no exit price implies.
• Borrow rate at entry and accrued financing at exit — plus the general-collateral or hard-to-borrow flag.
• Trigger price alongside fill price — both, or trigger-based slippage is unavailable forever.
• Expected R and thesis at entry — the decision-quality record, and the part no firm system holds.
• MAE and MFE — with a liquidity or ADV band, so the stop-placement cut is possible.

How SignalDeck Handles It, and What It Does Not

One part of this list SignalDeck does natively and, as far as we know, unusually. The concentration check in section 4 is built in and it runs before the trade, not after. When you enter a symbol on the New Trade form, your open equity book is grouped into correlated clusters by sector, and if the prospective position joins a cluster that is already heavy you are told so while the size is still editable — with the combined dollar, R and percent-of-account exposure for that cluster. The sector is resolved automatically from the ticker's SIC classification rather than typed by you, which matters because self-labelling is precisely where this check fails: the source's own worked example is UNP + NSC, both SIC 4011 railroads, flagged as one exposure. Where a ticker cannot be resolved, the position still counts toward total open risk but joins no cluster, and nothing silently guesses.

Also present: MAE and MFE per trade, trigger-based execution metrics (fill against your stop or limit trigger, limit penetration, price improvement, exit slippage), per-strategy aggregation with the positions each strategy is currently holding, and the R, expectancy, SQN and Monte Carlo layer underneath all of it.

Four gaps, stated plainly because this audience will find them in ten minutes anyway. Sector is not stored on a trade — it is resolved live for the open book, which is what makes the pre-trade check work, but it means there is no historical P&L-by-sector report; the section 2 table is something you would build from a tagged export today. There is no return-on-capital-deployed metric; the inputs are in the data but the calculation is yours. There are no borrow, locate or financing fields — those go in tags and notes, which is workable for the flag and awkward for the rate. And execution is benchmarked against your own trigger, never against arrival price or VWAP, because SignalDeck does not ingest a market data feed alongside your fills. Catalyst is a tag rather than a first-class field, which works well enough in practice but will not sort itself.

That is the honest boundary as of August 2026. If your firm's stack already answers the two questions in section 7, none of this is for you. If it does not, the pre-trade cluster check is the piece worth having on its own.