Tariff arithmetic: what the consensus
elasticity gets wrong.

Tariff modelling is a domain where the standard economic model, applied mechanically, consistently produces estimates that diverge from observed outcomes by 40 to 60%. The standard model takes the tariff rate, multiplies by the import share of consumption in the affected category, and produces a headline PCE impact. This calculation is not wrong in a directional sense. It is wrong in a quantitative sense because it assumes a stable pass-through elasticity that the evidence consistently shows is not stable. The desk's framework for understanding tariff arithmetic begins with this failure and works backward to the five variables that determine the actual elasticity in any specific tariff episode.

The desk's full five-variable model is presented in the dedicated tariff pass-through note. This Live Desk piece has a narrower focus: it explains specifically what the consensus tariff elasticity estimate gets wrong, quantifies the magnitude of the error, and describes the monitoring approach that the desk uses to update its elasticity estimate in real time as the relevant variables move.

The consensus elasticity estimate for the current tariff configuration, as represented by the median forecast in the Federal Reserve Bank of New York's survey of professional forecasters, is approximately 0.35 to 0.40 percentage points of core PCE per 10 percentage points of effective tariff rate, with a lag of approximately 3 to 6 months. The desk's own estimate is 0.25 to 0.28 percentage points of core PCE per 10 percentage points of effective tariff rate, with a lag of 6 to 9 months. The gap between the consensus and the desk's estimate is approximately 30 to 40% in magnitude and 3 months in timing. Both dimensions of the gap matter for monetary policy positioning.

The consensus overestimates the magnitude of tariff pass-through for four reasons that the desk quantifies separately and then aggregates.

The first reason is inventory buffer underestimation. The standard model assumes that tariff costs pass through immediately to consumer prices, but in practice the inventory buffer at the importer level delays the pass-through by the number of months of inventory on hand. The desk's current estimate of the inventory buffer for electronics (the largest single tariff-affected category) is 2.1 months, meaning that electronics importers can absorb approximately 2.1 months of tariff costs before they are forced to raise prices. The consensus model uses a 1.5-month buffer assumption, which is calibrated to a different tariff episode (2018 to 2019 Section 301 tariffs) when inventories were lower.

The second reason is retailer margin compression underestimation. The standard model assumes that retailers pass through 100% of import cost increases to consumers. In reality, retailers face a competitive dynamic that forces them to absorb a fraction of the cost increase through margin compression rather than price increases, because price increases reduce volume and volume loss is the variable retailers manage most carefully. The desk estimates that the current retail gross margin compression, at approximately 1.2 percentage points below the 2022 to 2023 average, has absorbed approximately 18 to 22% of the tariff cost that would otherwise have passed through to consumer prices. The consensus model assumes 0% retailer absorption.

The third reason is FX offset overestimation. The consensus model correctly identifies FX offset as a factor, but overestimates its magnitude. The USD has appreciated approximately 3.8% against the CNY, VND and MXN basket since the April 2025 tariff announcement. This appreciation reduces the dollar cost of imports from those countries, which partially offsets the tariff cost in dollar terms. But the consensus model applies the FX offset at the full 3.8% rate across all tariff-affected imports, when in fact the offset is only partial because many supply contracts were locked in at pre-appreciation exchange rates and the FX offset will not be fully realised until those contracts expire and renew.

The fourth reason is supplier substitution velocity overestimation. The consensus model credits supplier substitution, the shift of import sourcing from tariff-affected to tariff-exempt countries, with a 12-month impact that reduces the effective tariff rate by approximately 15 to 20 percentage points. The desk's tracking of actual import origin data shows that the substitution has been more partial than the standard model assumes: approximately 30 to 40% of the substitution volume represented by the data is the tariff routing phenomenon (goods assembled in Vietnam or Mexico from Chinese components, not genuine supply chain diversification), and this routing-based substitution does not reduce the underlying cost pressure in the same way that genuine substitution does.

Taken together, the four factors produce the desk's estimate that the actual magnitude of the tariff pass-through to core PCE is approximately 30 to 40% lower than the consensus estimate, for the same nominal tariff rate.

The consensus lag estimate of 3 to 6 months for tariff pass-through is derived primarily from the 2018 to 2019 tariff episode, specifically from the regression analysis of PCE components against the tariff implementation dates in that episode. The desk's view is that this calibration is inappropriate for the current cycle for two reasons.

The first reason is the difference in inventory dynamics. In 2018 to 2019, US importers were caught with relatively lean inventories when the tariffs were announced, because the supply chain management trend of that period was toward just-in-time inventory reduction. The lean inventory position meant that the tariff cost flow-through to retail prices was fast, producing the 3 to 6 month lag that the consensus has calibrated to. In the current cycle, importers entered the tariff period with above-average inventory levels, partly as a precautionary response to the supply chain disruptions of 2020 to 2022 and partly as a deliberate front-loading of inventory in anticipation of tariff escalation. The above-average inventory position delays the cost pass-through by approximately 2 to 3 months relative to the 2018 to 2019 calibration.

The second reason is the difference in retailer behaviour. In 2018 to 2019, retailers were early in the inventory front-loading cycle and had limited pricing power with consumers who were still adjusting to the tariff environment. In the current cycle, retailers have 5 years of experience with tariff pass-through, have developed pricing playbooks for managing the cost increase, and are more confident in implementing price increases because the consumer's experience with post-COVID inflation has created tolerance for price increases that did not exist in 2018 to 2019. This experience effect is partially offset by the higher starting price level, which reduces consumer tolerance for further increases, but on net the desk estimates that the experience effect adds approximately 0.5 to 1.0 months to the lag compared to the 2018 to 2019 episode.

The combined effect of the longer inventory buffer and the retailer experience adjustment is a tariff lag of approximately 6 to 9 months in the current cycle, versus the 3 to 6 month consensus assumption. This 3-month timing difference is consequential for monetary policy because it determines whether the Fed is easing into a period of peak tariff PCE impact or easing after the tariff impact has peaked. If the consensus lag of 3 to 6 months is correct, the tariff PCE impact peaked in Q1 2026 and the Fed is easing after the peak, which is benign. If the desk's lag of 6 to 9 months is correct, the tariff PCE impact will peak in Q2 to Q3 2026, and the Fed will be easing into the peak, which complicates the insurance cut narrative described in the LD-G10-03 note.

Because the tariff pass-through elasticity is not a constant but a function of five variables that each move over the course of the tariff implementation cycle, the desk has developed a real-time monitoring framework that updates the elasticity estimate weekly based on the most current data for each variable. The five variables and their data sources are:

Inventory buffer: The US Census Bureau publishes inventory-to-sales ratios monthly for 13 retail categories at a lag of approximately 5 to 6 weeks. The desk supplements this with weekly data from the Logistics Management Institute's inventory index and from the National Retail Federation's monthly surveys, which provide higher-frequency but less comprehensive inventory data. Current reading for electronics: 2.1 months, down from 2.5 months in Q3 2025.

Retailer margin position: Quarterly earnings data from the 15 largest US retail companies provides gross margin data with a lag of approximately 6 to 8 weeks. The desk supplements this with the Bureau of Economic Analysis's quarterly gross operating surplus data for the retail sector, which is more comprehensive but arrives with a longer lag. Current reading: 1.2 percentage points below the 2022 to 2023 average.

FX offset: Daily exchange rate data for the CNY, VND and MXN against the USD, combined with the import origin share data from monthly Census Bureau trade statistics, allows the desk to calculate the effective FX offset on a monthly basis. Current reading: 3.8% USD appreciation against the weighted import basket, generating approximately 7 to 9% tariff cost offset in dollar terms.

Supplier substitution: The desk's monthly routing purity analysis, described in the Vietnam/Mexico routing note, estimates the fraction of the apparent import origin shift that represents genuine substitution versus tariff routing. Current estimate: approximately 30 to 40% of the apparent substitution is genuine, 60 to 70% is routing-based.

Political timing: Weekly Congressional action monitoring, White House tariff policy communications tracking, and USTR Section 301 review calendar. The desk maintains a tariff implementation probability calendar that assigns probabilities to each announced tariff measure, categorised by whether it has entered the formal regulatory comment period, received a Congressional Review Act challenge, or been subject to an executive order delay.