The Death-Over Market Is Walking the Wrong Way: What 102 ILT20 Matches of Ball-by-Ball Data Reveal

Sharjah Cricket Stadium, one night last season. The 19th over. Six balls, five runs conceded. No boundary, no wide, just six deliveries that kept the batter guessing on line and length the whole way. Nobody in the stands stood up. The highlight package skipped that over. The fantasy leaderboard did not move the bowler's price by a single rupee. He did not play the next three matches.

In the same season, another bowler went at roughly 11 an over in the death phase and kept his place — because his wicket column looked healthy. The work of restricting runs does not photograph well, so the market pays less for it; the work that prints big on a scorecard gets paid more. My model suggests the true gap between those two contributions in overs 16 to 20 is roughly the inverse of the gap the market prices. The death-over price is being set on the wrong variable.

Context: where the data comes from

ILT20 began in January 2026 in the United Arab Emirates with six teams — Abu Dhabi Knight Riders, Desert Vipers, Dubai Capitals, Gulf Giants, MI Emirates and Sharjah Warriorz. Each season runs to 34 matches. Across three seasons that is 102 matches, roughly 24,000 deliveries. Gulf Giants won the 2026 title, MI Emirates took 2026, and Dubai Capitals won 2026 (source: ILT20 season records; the 2026 final was played on February 9, 2026).

I tagged those 24,000 balls one by one. Each delivery got an expected-runs value (xR) built from the phase of the innings, the batter's career T20 strike rate, the bowler's type, the venue and the scoreline. I then computed xRS — expected runs minus actual runs — per over. Yorker frequency I encoded by hand from broadcast pitch maps, because I do not hold Hawk-Eye grade ball-tracking data. That is the model's largest limitation.

Economy rate is unfit for valuation, and anyone who has watched the format knows why. It does not separate the quality of the batter — eight runs to a No. 7 is not eight runs to a top-order player. It does not separate match state — the 19th over of a chase already gone is not the 19th over of a five-run game. And the biggest hole: it never tells you the ratio of dot balls to wickets inside that over.

The database did not replace the game; it translated it. So my accounting starts not from basic economy but from separate prices for a boundary blocked and a wicket taken.

The evidence chain: what a wicket is worth versus what a dot ball is worth

In the 16-to-20 phase, one dot ball saves roughly 0.62 xR in my model. One wicket saves about 1.9 xR, because a wicket does more than waste a delivery — it brings a new batter in, breaks strike rotation, and resets the next over's plan. On that reading, wickets should dominate everything.

Frequency says otherwise. In the last five overs, an average over contains about 2.4 dot balls and 0.3 wickets. Run the arithmetic: roughly 72 percent of all runs saved in that phase come from dot balls, 28 percent from wickets. The deliveries that save the most runs are the ones valued least. In a sketched franchise fee database, death-over wicket-takers carry contracts about 40 percent larger than bowlers who generate dot pressure at similar xRS.

The yorker tax

Bowlers who land more than 30 percent of their last-five-over deliveries on yorker length tend to take fewer wickets — the batter cannot time it, so the ball squirts to mid-on rather than to a fielder. Those same deliveries cut boundary rate roughly in half. Scouts call these bowlers backup options; selection committees list them seventh or eighth in the attack. In the model they sit in the top three. That gap is the whole point: the eye does not see it, the scorecard does not record it, the model can count it as runs.

UAE names such as Muhammad Waseem, Junaid Siddique, Aayan Afzal Khan and Muhammad Jawadullah sit on squad sheets but are frequently absent from the model's most profitable column. Their season is ten matches long, with perhaps 24 to 36 death deliveries. That sample struggles to capture even the work of a death specialist like the USA's Ali Khan.

Match-state weighting: where captaincy evaluation goes wrong

When a game is already lost, captains often hand the best death bowler one more over on the hope of something happening. Boundaries follow, economy balloons, and the blame lands on the bowler. My dataset carries plenty of supposedly expensive death bowlers posting 13 or 14 an over in innings where the win probability was already below 10 percent. Without match-state weighting, you fuse pressure with bad luck. Once you do that, decision quality and outcome luck can no longer be separated.

This is where a decision memo earns its place. In 2026 I built an xR-based shortlist for a franchise. At the top sat a 24-year-old pacer with strong death-over xRS, a low wide count and high yorker frequency. The club signed a 34-year-old known name at roughly double the wage because it wanted experience. The veteran took five wickets in sixteen matches at an economy above ten, and the side slid down the table. Two separate questions live here — how bad the process was, and how bad the outcome was. Conflate them and next season you repeat the mistake with a different name.

The pattern hidden in negative space

A shot map is memory with coordinates. A wagon wheel tells you where a batter scores. In the death phase my interest runs the other way — the channel a bowler refuses to bowl into. A bowler whose sparse sample is dense with the wide yorker outside off tends to hold the most stable xRS, because the prior barely shifts on so little data. I found the dot-pressure bowler hiding in the negative space of a shot map. That silence reads as boring, and boring is precisely the asset.

Sample size: how much noise you are buying

During the empty-stadium months I scraped four years of records on 1,800 cricketers, and I learned that the silence of empty stadiums became my loudest dataset. The lesson travels. Death-over performance is an ultra-high-variance game. At a 24-ball death sample, Bayesian shrinkage runs around 60 to 70 percent — meaning roughly two-thirds of the price a team settles on comes from the prior, and one-third from observed performance. A franchise that finalises a contract off 12 balls and six dots is mostly buying its own soundbite.

The contrarian angle: correlation is not causation

My own model deserves suspicion here. A dot-ball specialist is sometimes a product of usage — the captain calls him when the opposition's seven and eight are at the crease, so the good number is an arrangement rather than a skill. A wicket may also do more than waste a delivery: it changes the next batter's plan, and part of that cascade my xR baseline still fails to price. On some questions the market may simply be right.

Above that sits unmodeled variance no dataset will ever hold — nerves, family, visas, injuries, the absence of a physio. Many associate bowlers in the Gulf franchise system arrive from environments with no written injury history. Throwing a man the 19th over of a big match is not the same as dropping him into a ten-year support system, and my database flattens the two.

The structure runs deeper into political economy. League money comes from the Indian broadcast market; crowds come to see familiar names. Buying the recognisable player is therefore the incentive-compatible decision. What I call oversight reads as correct business. The associate cricketer here is not a discarded option; he is a discarded incentive.

Takeaway: what I will track in the next ten matches

Across the next ten games I will log three numbers — xRS for every bowler with 30 or more death balls, yorker frequency, and how often a side drops a low-economy death bowler immediately after one wicketless spell. That raises a question for the ILT20 auction cycle: will prices start drifting toward xRS, or will the wicket column keep getting quantified further?

I do not predict transfers; I reconcile the lag between rumour and contract. The question now is what data is actually for, when the market cannot tell its own error apart from its own luck.