Reading the risk metrics on your Polymarket curve
An equity curve is more than a line going up or down. A few numbers next to it — period return, daily hit ratio, market win rate and Sharpe — say how good the ride was, not just where it ended. Here is what each one means and exactly how it is figured.
Period return
The return is your result over a window as a share of what you put in, so you can compare a small wallet to a big one. Pick the window on the chart — month to date, year to date, or since inception — and the return is measured over exactly that span. Two wallets can both be up 200 dollars; the one that did it on less money has the higher return.
Daily hit ratio — how often a day goes your way
The daily hit ratio is the share of days you finished in the green:
Flat days — where the result did not move — are left out, so a quiet holiday does not drag the number down. A daily hit ratio of 55% means that, over the window you chose, you closed the day up more often than down. On its own it says nothing about size: many small wins and one big loss can still be a losing wallet, which is why you read it next to the return.
Market win rate — how often your markets pay
The market win rate counts markets instead of days: of the markets you closed in the window, the share that ended in profit. A market's result is everything it made — sales, redemptions and what is still held at the payout price — and Yes and No of the same market count together, once, so a hedge is a single market. A market where you still hold a live position waits until it closes, and break-even markets are left out, like flat days. It is a count, not a dollar figure: read it next to the return.
Payoff, profit factor and expectancy — how big the wins are
The win rate says how often you win; these three say by how much, on the same closed markets. A payoff of 2× means the typical win is twice the typical loss, so you can be right less than half the time and still come out ahead. A profit factor above 1 means the wins paid for the losses. The expectancy is the average result per closed market in dollars, and below it, the total result over the money invested in those markets. With no losing market, payoff and profit factor show a dash.
Holding time — do you sit on your losers?
For the same closed markets, the holding time runs from your first entry to the market's close. The cards show the median for markets closed in profit and for those closed at a loss, with the 25%–75% range below, and the ratio between the two medians. Traders tend to sell winners early and hold losers hoping they come back — the disposition effect. A ratio of 2× means your typical loss stayed open twice as long as your typical win; well above 1×, it is worth asking whether you are cutting losers late.
Sharpe ratio — return per unit of risk
The Sharpe ratio rewards a steady climb and punishes a bumpy one. It takes your daily percentage results — each day's profit divided by the previous day's portfolio plus any deposit made that day — and compares their average to how much they swing:
Here each r is a daily return, n is the number of days, and √365 scales a daily figure up to a yearly one (the ACT/365 convention). This is an approximation of the textbook Sharpe: it assumes a risk-free rate of zero and uses the root-mean-square of the returns in place of the sample standard deviation. The two are close in practice, and the shortcut keeps the number stable on short windows. Loosely: above 1 is good, above 2 is excellent, and a negative Sharpe means the swings were not worth the return.
Calibration — where your edge is
The calibration chart groups your closed trades by entry price, in nine bands from under 15 to 85 and up (prices go from 0 to 100). For each band, the horizontal axis is the average price you paid and the vertical axis is what each share actually returned — 1 or 0 at resolution, the average sale price if you sold before, a blend of both if you did some of each:
The averages are weighted by the shares bought, so a 500-share trade counts more than a 5-share one, and the band's edge is exactly its total result divided by its total shares. The dashed diagonal is a fair price: a point above it means you bought cheaper than it turned out to be worth (edge), below it means you paid too much. The distance from the diagonal is the edge in points per share. Bigger circles have more trades, the vertical bar is a 95% range, and bands with fewer than five trades are faded — too few to read.
True Skill Score — your calibration in one number
The True Skill Score sums up the calibration chart in a single number. If x is the price you paid (0 to 1) and Y(x) is what each share actually returned at that price — the calibration curve — the score is:
Read the inside of the integral as a risk-adjusted edge, a small Sharpe ratio at each price level: the excess probability Y − x divided by the standard deviation of the event, √(x(1 − x)). A share at 50 swings a lot, so beating it by 2 points is ordinary; a share at 3 or at 97 barely swings, so the same 2 points are a much bigger feat. Dividing by the standard deviation gives the ends more weight, where beating the price is statistically harder, and it is also the weight that puts the sampling noise of every price level on the same footing. The factor 2/π pins the range to [−1, 1]: +1 means you always win, −1 that you always lose, and 0 a fair price on balance — no edge. The weight 1 ÷ (π √(x(1 − x))) is the arcsine distribution, so the score is also simply 2 × (the arcsine-weighted average of Y) − 1, and a constant edge c everywhere scores exactly 2c.
The dashboard integrates with the trapezoidal rule in the angle θ = arcsin√x, where the arcsine weight is uniform. Each calibration band with trades becomes a node (xᵢ, yᵢ): the average price paid and the average exit, weighted by shares, manual positions included. Two fixed nodes close the ends: x = 0 with Y(0) = 0 and x = 1 with Y(1) = 1, where the edge is zero. Each node weighs half the angle between its neighbors, which is its Voronoi cell in θ:
The masses, with those of the two ends, add up to 1. A calibrated curve (yᵢ = xᵢ) scores exactly 0, a single band scores its own edge (its mass is ½), and nothing blows up near 0 or 1. A band with no trade is not a node: the trapezoid passes over it. Bands with fewer than five trades count, and the card says how many. Because the price is measured at the node and not inside the cell, a few unusual curves sum slightly past ±1; the card shows the score capped at ±1 and the uncapped sum in its tooltip. Hovering over the card shows the three bands that weighed the most.
The averages yᵢ come from a finite number of trades, so part of each band's edge is luck. The True Skill Score shrinks each edge toward the fair price in proportion to that noise, as if every band also had k fictitious trades at exactly the price paid:
nᵢ is the band's effective number of trades: with share weights w, (Σw)² ÷ Σw², so one huge trade does not count as many. With 3 trades λ is 0.23 and an edge of +0.30 counts as +0.07; with 40 trades λ is 0.80; with 300, 0.97. Without skill, the risk-adjusted edge of every band has noise 1/nᵢ whatever the price, so the same k serves all of them (James–Stein shrinkage). k = 10 is deliberately conservative: showing as skill what was luck is the costliest mistake for an index called "True". A calibrated curve still scores exactly 0, each band keeps the sign of its edge, and the tooltip shows the λ of the three bands that weighed the most.
The badge. Over the whole history (no time decay, k = 10), at least 1,000 closed trades and a True Skill Score of 0.05 or more earn the "prices well, consistently" badge on the card. The 0.05 cut leaves the badge to roughly the top 5% of the skill ranking. The badge measures the quality of the price paid; it does not promise profit.
Two limits. The score summarizes, it does not characterize: a curve that is too good on longshots and equally bad on favorites can net out to 0, so look at the chart too. And it is not a proper scoring rule like the Brier score, which grades each forecast; it is an after-the-fact summary of results, like a realized Sharpe.
Result by category — where the money came from
The category table crosses months with the categories of your own classification (the same as the consolidation). Each closed market counts in the month it closed, with its whole result, so you can see which subjects carried you and when. The current month also adds the unrealized result of open positions at today's mark, and is labeled partial. Cells are tinted like the daily heatmap; hover for the ROI and the number of markets.
Manual positions — the ones you enter by hand for trades made outside Polymarket — count in every metric above, like any other position; those entered as a dollar value, with no per-share price or exit date, count only in the result by category (in the current month) and in concentration.
Daily VaR and expected shortfall — what a bad day can cost
The 95% daily VaR is the loss that was exceeded on only 5% of days. The dashboard gets it by historical simulation: it takes the daily returns of your own curve (the same ones the Sharpe uses, as a % of the previous day's equity and without deposits), keeps up to the last 365 days and applies them to today's equity.
Expected shortfall answers the next question: when the day is one of the bad ones, how much do you lose on average? It is always equal to or larger than the VaR and says more about the tail. Both need at least 60 days of curve; before that the card shows how many days are missing. Manual adjustments are included, because they are part of the curve. One limit: the simulation only knows the days you have already lived. A wallet that changed a lot in size or in what it holds can carry a risk today that is quite different from what its history shows.
Effective positions — how many positions you really have
The share is the value of the open positions in one event divided by the value of all of them, at the mid of the book. Positions in the same event count as a single position, and manual positions are included. Four events of the same size give 4; ten events where one holds 90% of the value give close to 1. It is a snapshot of now and does not follow the chart window. Below 10 effective positions the card warns that the wallet is concentrated: with so few positions the daily result does not follow a normal distribution, the historical VaR tends to understate the risk, and the number to look at is the resolution risk.
Resolution risk — what if everything resolved now
A prediction market ends at 1 or 0, and daily swings do not show that. The dashboard draws 5,000 scenarios in which each open position pays 1 or 0 with the probability its price implies, and shows the loss in the worst 5% of scenarios and the chance of losing more than 10% and more than 25% of your equity. How the draw works, step by step, is in the tutorial.
Liquidity risk — what it costs to get out
The wallet is valued at the mid of the book, but nobody sells at the mid. The dashboard simulates selling everything at once, at market: each position walks down the bids of its own market for the full size, and the gap to the mid value is the exit cost.
What the book cannot absorb counts as zero, the worst case, and is shown separately as "not enough liquidity to fully exit". The card shows the cost in dollars and as a % of equity, the equity that would be left on exit, and the positions that cost the most to leave. Manual positions entered as a dollar value are left out; the ones entered as a position use the book of their reference market. It is a snapshot of the book right now: in a thinly traded market it changes fast.
Why the numbers move with the window
All of them depend on the span you select. A great month can carry a strong Sharpe that fades once a rough week enters the year-to-date view. A handful of days is too few to trust a Sharpe at all, so treat short windows as a hint, not a verdict — the more days in the window, the more the numbers mean.
Read your wallet like a fund.
Daily equity curve, IRR, Sharpe, drawdown and alerts on your phone.