Kalkas

The record · Horse racing

Horse racing: calibration and closing-line value

Do the probabilities mean what they say, and does the price taken beat the close? Both charts are drawn from the resolved-call ledger by the same rules as the scoreboard, with the sample size beside them.

Reliability diagram

Calls are grouped by the probability they stated. A calibrated forecaster’s dots sit on the diagonal; a bar that misses it is a claim the sample does not yet support. Bins with few calls have wide bars.

aggregateracing.win

n = 5,554 scored calls

Reliability diagram for aggregate (racing.win)9 probability bins over 5,554 scored calls. Each dot is the share that came true in its bin, with a 95% Wilson interval bar, against a diagonal for perfect calibration.0%0%20%20%40%40%60%60%80%80%100%100%Stated probabilityCame true
Dot: how often calls in the bin came true (its size grows with the count). Bar: 95% Wilson interval. Dashed diagonal: perfect calibration.
Show the bins as a table
Reliability diagram for aggregate (racing.win)
Stated probabilityCallsMean statedCame true95% interval
0% to 10%3,8380.0420.0420.036 to 0.048
10% to 20%1,1560.1430.1510.131 to 0.172
20% to 30%3230.2420.2410.198 to 0.291
30% to 40%1230.3400.3980.316 to 0.487
40% to 50%590.4440.4580.337 to 0.583
50% to 60%290.5460.6550.473 to 0.801
60% to 70%180.6400.8890.672 to 0.969
70% to 80%70.7440.8570.487 to 0.974
80% to 90%10.8210.0000.000 to 0.793

racing-marketracing.winmarket baseline

n = 5,554 scored calls

Reliability diagram for racing-market (racing.win)9 probability bins over 5,554 scored calls. Each dot is the share that came true in its bin, with a 95% Wilson interval bar, against a diagonal for perfect calibration.0%0%20%20%40%40%60%60%80%80%100%100%Stated probabilityCame true
Dot: how often calls in the bin came true (its size grows with the count). Bar: 95% Wilson interval. Dashed diagonal: perfect calibration.
Show the bins as a table
Reliability diagram for racing-market (racing.win)
Stated probabilityCallsMean statedCame true95% interval
0% to 10%3,8380.0420.0420.036 to 0.048
10% to 20%1,1560.1430.1510.131 to 0.172
20% to 30%3230.2420.2410.198 to 0.291
30% to 40%1230.3400.3980.316 to 0.487
40% to 50%590.4440.4580.337 to 0.583
50% to 60%290.5460.6550.473 to 0.801
60% to 70%180.6400.8890.672 to 0.969
70% to 80%70.7440.8570.487 to 0.974
80% to 90%10.8210.0000.000 to 0.793

aggregateracing.win.jumps

n = 2,523 scored calls

Reliability diagram for aggregate (racing.win.jumps)9 probability bins over 2,523 scored calls. Each dot is the share that came true in its bin, with a 95% Wilson interval bar, against a diagonal for perfect calibration.0%0%20%20%40%40%60%60%80%80%100%100%Stated probabilityCame true
Dot: how often calls in the bin came true (its size grows with the count). Bar: 95% Wilson interval. Dashed diagonal: perfect calibration.
Show the bins as a table
Reliability diagram for aggregate (racing.win.jumps)
Stated probabilityCallsMean statedCame true95% interval
0% to 10%1,4760.0400.0470.037 to 0.059
10% to 20%5700.1450.1510.124 to 0.183
20% to 30%2440.2430.2580.207 to 0.317
30% to 40%1370.3450.3720.296 to 0.456
40% to 50%520.4480.3270.215 to 0.462
50% to 60%280.5480.5000.326 to 0.674
60% to 70%110.6470.7270.434 to 0.903
70% to 80%40.7541.0000.510 to 1.000
80% to 90%10.8681.0000.207 to 1.000

racing-marketracing.win.jumpsmarket baseline

n = 2,523 scored calls

Reliability diagram for racing-market (racing.win.jumps)9 probability bins over 2,523 scored calls. Each dot is the share that came true in its bin, with a 95% Wilson interval bar, against a diagonal for perfect calibration.0%0%20%20%40%40%60%60%80%80%100%100%Stated probabilityCame true
Dot: how often calls in the bin came true (its size grows with the count). Bar: 95% Wilson interval. Dashed diagonal: perfect calibration.
Show the bins as a table
Reliability diagram for racing-market (racing.win.jumps)
Stated probabilityCallsMean statedCame true95% interval
0% to 10%1,4760.0400.0470.037 to 0.059
10% to 20%5700.1450.1510.124 to 0.183
20% to 30%2440.2430.2580.207 to 0.317
30% to 40%1370.3450.3720.296 to 0.456
40% to 50%520.4480.3270.215 to 0.462
50% to 60%280.5480.5000.326 to 0.674
60% to 70%110.6470.7270.434 to 0.903
70% to 80%40.7541.0000.510 to 1.000
80% to 90%10.8681.0000.207 to 1.000

Closing-line value over time

Each act compares the price taken with the closing price observed after it. Above zero means the price taken was better than the close. Acts with no later closing price are not plotted.

n = 5 acts · Mean closing-line value +25.9%

Not enough resolved acts yet (n = 5)

Closing-line value needs at least 30 acts with a closing price observed after the decision.

Informational only, not advice and not an invitation to trade or bet. Everything runs on paper. Past results do not predict future ones.