Death-Overs Economy Doesn't Repeat — So Why Does the Auction Keep Buying It?

In the summer of 2026, in Hyderabad, three months before my statistics finals, I hand-coded 4,200 IPL deliveries from a laggy broadcast feed into a spreadsheet nobody had asked for. The output was a single thread on Bhuvneshwar Kumar's Purple Cap season — 26 wickets — showing that 78% of his overs 17–20 deliveries landed in the yorker band, and that his 7.02 death economy was structural, not streaky. That thread drew 400,000 impressions and an internship offer from a Hyderabad sports-data vendor.

Seven years later I picked up the same coding spec and re-ran it across several seasons with one question: does that 2026 conclusion repeat? The answer made me uncomfortable. Among bowlers who delivered more than 200 death balls in two consecutive seasons, the year-on-year correlation of death economy came out at roughly 0.3. The 95% confidence interval on that estimate touched zero. Which means the thing franchises pay crores for is something I could not, from my own data, measure with confidence. That is what this piece is about.

The auction is a public pricing experiment, and it has a settlement date

Franchise scouting has quietly replaced the zonal selector over the last fifteen years, and it has a side effect nobody discusses. The IPL mega auction is one of the few markets where people publish their bets. Every buy is a hypothesis with a settlement date. My job is to go back two seasons later and grade the room.

November 24 and 25, 2026, Jeddah. The biggest deal of that auction was Rishabh Pant — ₹27 crore to Lucknow Super Giants, then the highest price paid at an IPL auction. Behind him: Shreyas Iyer (Punjab Kings, ₹26.75 crore) and Venkatesh Iyer (Kolkata Knight Riders, ₹23.75 crore). All three batters. The headlines went to batting. But what shapes a squad for three seasons is the second-order spending: bowling, specifically overs 17 to 20.

Every franchise is solving the same problem — the fifth bowler. If your fifth bowler is a part-timer, your specialist gets squeezed into overs 14 to 16 and his death numbers vanish. That is a team-design problem, not a bowler problem. The market prices it as a bowler problem.

My coding spec, because a dashboard never sends its spec

Before I look at any economy figure I write down what I counted. Without a settled definition there is no argument to have.

• Death overs = overs 17, 18, 19, 20 of an innings. Powerplay and middle are separate buckets.
• Wides and no-balls were logged into the ball count, because they count in economy; an economy built on legal deliveries only is incomplete to me.
• Yorker band = a delivery whose pitch point falls within roughly 1.5 metres of the crease. That is my definition, not a provider's; I coded it frame by frame and dropped the ball whenever the broadcast cut away.
• Dropped catches did not change wicket attribution, but I logged them in a separate column so a bowler's luck and his bowling stayed apart.

I hand-coded four thousand two hundred balls before I trusted a dashboard. I did it because I needed to know what someone means when they say "his death economy is nine." The answer: he is averaging four different jobs.

Raw death economy has the match situation dissolved into it

I split every death ball into two buckets — low pressure (required rate under 9) and high pressure (required rate above 12). The same bowler's number moves between them, and it moves more than most bowlers differ from each other: in my sample the average gap was about 1.8 runs per over.

The bigger finding is that the direction of that gap was not uniform — in roughly a third of cases the sign flipped. For some bowlers economy rose under pressure, for others it fell. Which is the root of the problem: if high-pressure death and low-pressure death are two different jobs, an average of the two cannot be a price.

I learned that lesson expensively in 2026. Sixty-four matches gave me one xG model and a finishing lesson. The model was clean; the finishing was not. Porting shot-weighting into cricket meant translating every imported term — not "expected goal" but "expected runs per delivery by line and length," with the match requirement sitting as its own variable rather than dissolved into shot quality.

Take the case: one bowler defends 30, another 55. In the first, the batter's risk is compulsory; in the second he chooses it. Dropping both into one column and saying "8.9, decent" is exactly where the broadcast graphic and the coach's need part ways.

Who the ball was bowled to is a bigger variable than economy

Before I rank a player, I ask what the ball was doing. At the death that question always lands where the conversation doesn't go: who was the ball being bowled to?

Death overs are played in two completely different states. Defending, you meet numbers 5–7, finishers who arrived two overs ago and have had no time to set up. Chasing, you meet the top order that has faced 30 balls, read the pitch, and worked out the six-an-over arithmetic. In my coded set, a set top-order batter's death-overs strike rate ran consistently above a finisher's on the same deliveries — and the gap between teams was smaller than that gap.

Which means a bowler's death economy is substantially an output of his captain's plan, not his own bowling. The man bowling overs 19–20 defending is fighting a hitting finisher on a flat deck. The man bowling over 17 chasing is fighting a set batter. Their numbers sit side by side in the same public table without saying they weren't doing the same job.

And something twenty years of watching from a gallery where you can see the ball pitch has taught me: what changes most at the death isn't the track map, it's the batter's posture. Same bowler, same pitch, two different batters, and his two overs can go for 4 and 14. I learned to count what the broadcast camera leaves out.

The fifth-bowler problem: architecture, not personnel

Suppose a side has six bowling options and one is a part-timer. The plan has to spend that part-timer somewhere in overs 12–15. Now watch the specialist: he gets pushed to over 16, and overs 19–20 stop being his. His death-economy column empties for the season. Next year he moves to a side that uses him at 19–20, and suddenly he has death numbers again.

That is a role change, not a talent change. The market buys it as talent.

The 2026 IPL in the UAE made this visible. Empty stadiums made dew a variable I could no longer ignore — in Dubai, spin economy climbed from 6.9 to 8.4 after the 16th over of a chase. The death-overs collapse everyone attributed to spinners that season belonged to a wet ball. When Mitchell Marsh went down with an ankle injury in the season's first match, leaving no seam-bowling replacement, we pushed hard for Jason Holder — because the problem was rebalancing roles, not finding a better bowler.

The sample arithmetic nobody printed on the auction table

A death specialist bowls about three death overs across fourteen league games — roughly 42 overs, or 252 balls. Take a per-ball standard deviation around 1.6 and a mean near 1.55 runs per ball (9.3 an over). The standard error of that mean over 252 balls is about 0.10 runs per ball, or 0.6 runs per over. A 95% interval of roughly ±1.2 runs per over.

Read that carefully: even after a full season, two bowlers with death economies of 8.4 and 9.4 are not statistically distinguishable. This is fourteen matches of data, not a small sample — and still not distinguishable.

Now the part that cuts against my own thesis, which I would rather publish than hide. Bowlers with 200-plus death balls in two consecutive seasons gave me about twenty pairs. A correlation of 0.3 on twenty pairs has a 95% interval stretching from roughly zero to 0.65. So I cannot say death economy does not repeat. I can only say I don't have the sample to settle it, and the market is behaving as though it does. Where we don't know, we write "we do not know yet."

What does repeat: where the ball pitched, not how many it cost

Economy is an outcome metric — batter's shot, dropped catch, fielder's position, match state all dissolved into it. Its variance is large and the sample runs out. Yorker-band percentage is a process metric — binary, per ball, under the bowler's control. Over 252 balls, a bowler landing 70% in the band has a standard error near 2.9 percentage points, an interval of about ±5.7. One season tells me whether a bowler can bowl a yorker. One season does not tell me whether he will be cheap at the death.

Bhuvneshwar's 2026 was exactly this: 78% in the band, and that is why I called the 7.02 structural rather than streaky. But when I tested whether yorker-band percentage predicts death economy across bowlers, the relationship existed and the slope was modest — because a missed yorker at the death is punished far more than a good-length ball. A bowler landing 70% and missing badly on 30% can average worse than one landing 55% and falling back on a safe length.

So the right question is not "what is his death economy." It is "what is his yorker-band percentage, and what is his backup line when he misses" — the skill that repeats, plus the plan that covers the miss. My job is to make the model legible to people who must act, and a coach cannot act on 8.9. He can act on a band.

The boring explanations first, then the inversion

There is a charge against me, and it is fair: being right against the room built my reputation, and a reputation becomes a reflex. So here are the explanations that would make this whole story boring.

One: maybe the correlation is low because bowlers change — team, home ground, and a dew-drenched venue rewrites an economy. I isolated bowlers who stayed at the same franchise in the same role across both seasons. The correlation rose, but not to a level that justifies a ten-crore premium. Two: maybe the batter really does determine death economy, so I shouldn't be looking for bowler stability at all. That explanation strengthens my argument — if the outcome belongs to the batter, pricing the bowler's name is the error. Three: league-wide scoring inflation lifted everyone and flattened the correlation. I normalised each season against that season's league-average death economy. The picture held.

Here is the blind spot I want to point at: franchises buy a category called "death overs," then discover the category doesn't travel, because the actual skill isn't death bowling — it's bowling to a specific batter in a specific required-rate state. Those are different things, and the second has no public price.

And I pre-register the hypothesis so I can't move it later: if a larger sample, holding role constant, puts the year-on-year correlation of death economy above 0.6, I am wrong and should retract this piece. That is why I don't discard economy entirely. I only say the sample we buy it on is not worth the price.

What to watch next round

Next time an auction table is in front of you, delete the death-economy column and put three in its place: yorker-band percentage, high-pressure death economy (required rate above 12), and low-pressure death economy. If a bowler's gap between those two economies exceeds two runs an over, you are holding either a very expensive misunderstanding or a very cheap opportunity — and you are entitled to know which. What I don't know yet, and want to know after two more seasons of data: whether death bowling in India is a teachable skill, or merely the role somebody's team happened to leave open. Until I have that answer, I will not buy a bowler on his death number.