Who got Georgia's 29 cents?

Georgia's pump price for regular gasoline averaged 352.1¢ a gallon during its 2022 gas-tax holiday, 59.3¢ above the 292.8¢ it averaged in the 62 weeks before. Taken at face value, suspending a 29.1¢ tax raised prices by twice the tax. The comparison is misleading because gasoline rose everywhere that spring: across the 34 states with no holiday, the same two periods differ by 92.0¢. This case builds a better comparison in three steps and shows how far the answer moves with each choice.

On Friday, March 18, 2022, three weeks after Russia invaded Ukraine, Georgia suspended its state excise tax on motor fuel, 29.1¢ a gallon for gasoline. The suspension was extended several times and ran through January 10, 2023. Georgia suspended the tax again from September 13 to November 29, 2023, when it had risen to 31.2¢.

Supporters said the savings would show up at the pump. Stations and wholesalers could instead keep part of the cut. The share of a tax change that reaches the price drivers pay is called pass-through:Data. Weekly average price of regular gasoline for 39 states, January 4, 2021 to June 24, 2024 (182 weeks), from about 7,700 convenience-store fuel stations. The price is revenue divided by gallons, so it averages cash and credit prices. Twelve states with fewer than 30 reporting stations are left out.

If prices fell by the full 29.1¢, pass-through is 100%. The hard part is the phrase caused by the holiday.

Every estimate needs a counterfactual

A causal effect compares what happened with what would have happened without the policy:

Georgia's actual price P^GA_t is observed every week. The second term never is. Each method below is a different estimate of it, and each rests on a different assumption.

Before and after: prices rose 59 cents

The simplest counterfactual assumes Georgia's price would have stayed where it was. The estimate is the average price during the holiday minus the average before it:

Figure 1 shows why this fails. Georgia's price rose and fell with every other state's through the 2022 oil shock. A before-and-after comparison works only when nothing else is moving, and in a fuel market something always is.

Difference-in-differences: the answer depends on the comparison group

A better counterfactual borrows the change from places without a holiday. Difference-in-differences subtracts the comparison group's change from Georgia's:

The assumption, called parallel trends, is that Georgia's price would have moved like the comparison group's without the holiday. Two comparison groups are available without any modeling: Georgia's four neighbors, and the average of all 34 states that had no holiday.⁴

Against its neighbors, Georgia's price rose 19.9¢ less, a pass-through of 19.9 / 29.1 = 68%. Against all 34 states it rose 32.7¢ less, or 112%, more than the whole tax. A pass-through above 100% means stations cut prices by more than the tax, which a retailer has no reason to do. It points at the comparison group: the 34-state average includes regions whose 2022 prices followed their own refinery and supply problems, the West Coast above all, so it overstates what Georgia would have paid.

Synthetic control: a comparison group chosen by fit

Synthetic control builds the comparison group from the data. It finds the weighted average of other states that best reproduces Georgia's price before the holiday and uses that synthetic Georgia as the counterfactual during it.¹² The weights are never negative and add to one, and they minimize the squared pre-holiday gap:

The effect in any week is the gap between Georgia and its synthetic twin. Pass-through is the average gap during the holiday divided by the tax:

The donor pool is the 34 states with no holiday of their own. The weights concentrate in the Southeast: Alabama, South Carolina, North Carolina and Tennessee carry 0.67 between them, and Texas and Louisiana, the Gulf Coast states that supply the region by pipeline, add another 0.18. Over the 62 pre-holiday weeks synthetic Georgia tracks the real one with a root mean squared error of 1.3¢ a gallon.

During holiday 1 Georgia's price averaged 24.6¢ below its synthetic twin, a pass-through of 24.6 / 29.1 = 0.85.

Three checks that the gap is the holiday

The gap follows the tax calendar

In the first full week after the tax returned (January 16, 2023) the gap was +0.9¢. During holiday 2 it reopened at −25.1¢, 80% of the 31.2¢ tax, and it closed again when that holiday ended. A coincidence would have to follow the legislature's calendar twice.

No placebo state looks like Georgia

Pretend each donor state had the holiday, build its own synthetic twin from the other donors, and measure how much its gap grows after March 2022 relative to how well it fit before. Georgia's ratio, 19.6, is the largest of the 35 states; the next largest is 6.4. If the holiday did nothing, the chance that Georgia ranks first among 35 is 1 in 35, about 0.03.

No effect appears before the holiday

Pretend the holiday began in September 2021, six months early, and fit the weights on the weeks before that date. The estimated effect between the fake and the real start is +0.2¢.

The cut arrived slowly and the tax came back at once

The gap also shows how fast prices moved. In the first full week of holiday 1 the gap was −16.3¢, 56% of the tax; it reached −22.4¢ (77%) in week 4 and −23.9¢ (82%) in week 6. When the tax returned, the gap closed within a week. Fuel prices are known to rise faster after a cost increase than they fall after a decrease;⁶ here the asymmetry appears on a cost change set by law. Holiday 2 behaved differently: the gap was already −24.8¢ in its first full week.

Change the donor pool and the estimate moves

The 85% above depends on the states allowed into the donor pool. The explorer below refits synthetic Georgia on any set of donors you choose, using the same least-squares rule as the figures above, and also shows the simple average of the same states for comparison. Differences between the two methods are largest when the donors are a poor match for Georgia.

Run on subsets, the synthetic estimate is 78% to 92% for every pool that includes Southern states: the four neighbors alone (78%), the South alone (83%), all 34 donors (85%) and all but the neighbors (92%). Pools outside the South give 126% (all 22 non-Southern donors) and 112% (the Midwest). The West alone fits Georgia's pre-holiday price poorly, with an error of 11.3¢, and gives 76%. Simple averages of the same pools range from 68% (neighbors) to 161% (West).

What the holiday did to prices and to volume

About 24.6¢ of the 29.1¢ cut reached the pump during holiday 1, so 4.5¢ a gallon stayed in the supply chain. The state gave up the full 29.1¢ on every gallon sold, including the gallons it would have sold anyway. The holiday was a transfer to drivers with a leak of about 15%, larger in the first weeks. Studies of fuel-tax incidence find that most of a tax change reaches pump prices, with the share varying with supply conditions.⁵

The same method applied to gallons sold finds no detectable change statewide. With each state's gallons indexed to its own pre-holiday average, Georgia ran 3.0 points above its synthetic twin during holiday 1, against a standard deviation of 3.9 points across the placebo states. Station-level work on the same data, not included in the student files, shows where the extra buying happened: stations within 10 km of Georgia in neighboring states sold 6 to 8% fewer gallons during the holidays, as border drivers crossed the line. Case 5 takes up that border response.

What these data cannot show

Georgia's neighbors carry two-thirds of the weight in synthetic Georgia. If stations just across the line cut their own prices to keep customers, synthetic Georgia is pulled down and pass-through is understated. The station-level work finds neighbors' prices moved by less than 3¢ in either holiday, so the bias is small, but these state-week averages cannot rule it out.

The prices come from convenience-store fuel stations that report full weeks, mostly in the South and Midwest, and the weekly gap is noisy: through the summer and fall of 2022 it swung between −15¢ and −34¢ as prices spiked and fell. Trust the average over many weeks, the switching pattern across two holidays and the placebo ranking, and not any single week. The placebo p-value cannot go below 1/35 with 35 units. Georgia's tax is also indexed each January (29.1¢ in 2022, 31.2¢ in 2023), and several donor states index theirs, so gaps of a few cents outside the holidays are expected.

Questions for discussion

1. The same data gave +59¢, −20¢, −33¢ and −25¢. For each, state the assumption about Georgia's price without the holiday and say what in Figure 1 or Figure 2 makes it plausible or not.
2. A pass-through of 112% is impossible for a profit-seeking retailer. What does it tell you about the comparison group, and how would you check?
3. In the explorer, remove Alabama, South Carolina, North Carolina and Tennessee. The pre-holiday fit worsens only slightly. Why might an analyst still prefer to keep them?
4. The gap closed in a week when the tax returned and opened over a month when it was cut. If you ran a chain of stations in Georgia, what would you do with prices in each case, and what would a regulator conclude from watching you?
5. The state gave up 29.1¢ on every gallon and drivers kept 24.6¢. Who else gained or lost, and what would you need to measure to say?
6. Placebo tests give a p-value of at best 1/35. How many donor states would you want before you called a result significant at 1%?

@misc{singh2026gas,
  author = {Singh, Vishal},
  title  = {Who got Georgia's 29 cents?},
  year   = {2026},
  note   = {Teaching case, NYU Stern School of Business},
  url    = {https://vishalsingh.org}
}