best lead in each 36-month window since Jan 2016: same month in 2 (Mar 2026–Apr 2026), gas leads 1m in 126 (most windows), gas leads 2m in 1 (Apr 2023) of 129 windows.
lag-1 rule's call for Oct 2026 CPI: gas moved +0.02% in Sep 2026 → CPI MoM above Sep 2026's +0.65%. naive repeat says above +0.65% (Sep 2026's CPI % change was bigger than Aug 2026's). scored when BLS prints Oct 2026 CPI.
Oct 2026 gas so far: 3.0985 on 2026-10-19 (-9.87% vs Sep 2026 close). month not finished, so it's left out of every number on this page.
| since 2016, correlation | same month | gas leads 1m | gas leads 2m | gas leads 3m |
|---|---|---|---|---|
| monthly % (month-end close) | 0.22 n127 | 0.51 n127 | 0.19 n127 | -0.05 n127 |
| monthly % (month-average gas) | 0.51 n127 | 0.43 n127 | 0.11 n127 | -0.07 n127 |
| 4-month % (Goat's tile) | 0.40 n127 | 0.52 n127 | 0.43 n127 | 0.29 n127 |
| year-over-year % | 0.49 n128 | 0.58 n128 | 0.61 n128 | 0.62 n128 |
| hit rate, direction call | since 2016 | last 36m | last 12m |
|---|---|---|---|
| gas, last month (lag 1) | 61.1% 77/126 | 60.6% | 66.7% |
| gas, same month (lag 0) | 57.1% 72/126 | 63.6% | 77.8% |
| naive: CPI repeats last month | 41.6% 52/125 | 46.9% | n/a (8 of 12 months scoreable) |
| naive flip: CPI reverses last month (= 100 − repeat) | 58.4% 73/125 | 53.1% | n/a (8 of 12 months scoreable) |
| coin flip | 50% | 50% | 50% |
watching Tom's idea that gasoline leads CPI. it's a running check on past data, not a forecast.
method: CPI = FRED CPIAUCSL (CPI-U all items, seasonally adjusted), monthly % = index ÷ prior month − 1. gas = NYMEX RBOB front-month (yahoo RB=F), last daily close of each month, monthly % = close ÷ prior month's close − 1 (not seasonally adjusted). lead k = gas change k months before the CPI month. rolling window = 36 CPI months, needs 30+ valid pairs. direction call: CPI 'up' = this month's CPI % change is bigger than last month's, 'down' = smaller. gas call = sign of gas's monthly change (lag 1 = the month before; lag 0 = the same month, which is still known ~2 weeks before BLS prints). naive = CPI's direction repeats last month's; CPI's direction flips more often than not, so the reverse rule ('naive flip') is shown too. ties and missing months are skipped, never filled. plain-read rule: compare lag-1 rolling correlation now vs 12 months earlier; more than ±0.05 = stronger/weaker, otherwise about the same. gaps: BLS published no CPI for Oct 2025 (shutdown), so the changes that need it are blank.
rebuilt by refresh.py