Conformal Prediction and Distribution-Free Intervals: Finite-Sample Coverage Guarantees for Order Counts and Order Values: chapter answers
Shop owners widely believe that the mean plus or minus one point nine six standard deviations is a 95 per cent prediction interval. This book tests that belief with exact calculation. On thirty heavy-tailed past values the average two-sided coverage of the normal interval is only slightly below nominal, but the right-tail miss rate rises from a nominal 0.025 to 0.0585, and a one-sided 95 per cent stocking bound misses 0.0731 of the time; with an infinite sample the right-tail miss rate is still 0.0382, so a wrong shape does not fade as data accumulate. Split conformal prediction guarantees fi…
DOI pending · monograph page goes live after publication · 5DL-B-306 · 中文版
- Chapter 1 · The Coverage Gap of Normal Intervals: Heavy-Tailed Orders Push Every Miss to the Right
The Pareto rows tell the same story. With tail index 2.1, where the variance barely exists, two-sided coverage is 0.9429, the right-tail miss rate is 0.0571 and the one-sided bound misses 0.0690 of the time. Note the la… - Chapter 2 · The Correct Rank: Why "Plus One" Matters and Where the Finite-Sample Guarantee Comes From
Chapter 2 proves that the rank must be the calibration size plus one, times one minus alpha, rounded up; without the plus one, twenty calibration values cover only 0.9048. - Chapter 3 · The Price of the Guarantee: Interval Width, the Shape of the Score and the Extra Paid When Normality Holds
In the log-normal row with parameter one half, the oracle width is 2.289, the median width of the normal interval 2.367 and that of the conformal interval 2.960. In the log-normal row with parameter one, the oracle is 6… - Chapter 4 · How Large a Calibration Set: Coverage Given the Calibration Set Follows a Beta Law
Chapter 4 shows that coverage given a calibration set follows a Beta law; with twenty calibration values the chance that true coverage is below 0.90 is 0.1216. - Chapter 5 · Marginal Is Not Conditional: Traffic Bands, Customer Segments and Intervals Too Wide to Use
Chapter 5 separates marginal from conditional coverage: high-traffic days can fall to 0.8798, and a score scaled by traffic brings all three bands back near 0.95. - Chapter 6 · When Exchangeability Breaks: Weekend Effects, Growth Drift, Promotion Shocks and Three Repairs
Chapter 6 shows how weekend effects, growth drift and promotion shocks break exchangeability: pooled calibration covers Saturdays only 0.8252 of the time, a frozen calibration set misses 0.3602 of the time under growth,… - Chapter 7 · Classical Sources East and West: Guo Shoujing's "The Calendar Rests on Measurement" and Neyman's Confidence Intervals
Chapter 7 examines Guo Shoujing's principle that the calendar rests on measurement and Neyman's confidence intervals of 1937, and shows that a confidence interval for the mean, used as a prediction interval, covers next… - Chapter 8 · Closing the Loop: the Weekly Order-Count Interval Sheet, the Conformal Interval Calculator and a Pre-Registered Test
Chapter 8 gives a weekly order-count interval sheet, the distilled Conformal Interval Calculator and a pre-registered test; stocking to a one-sided 95 per cent bound, the normal bound runs out 0.0657 of the time and the…
Illustrative model calculation, not a live quote