Some Evidence on the Importance of Sticky Prices
📄 Summarized from the full manuscript · Human-reviewed for faithfulness before publication
In brief
How often do prices actually change, and does that match what economists' models of "sticky prices" assume? Using unpublished government data on 350 everyday goods and services, this 2004 paper finds prices change far more often than earlier studies suggested -- a typical price lasts about four months, not a year. But when the authors build those frequencies into the standard sticky-price models used to explain business cycles, the models predict inflation that is far smoother and more persistent than what actually happens in the data, especially for goods whose prices change least often.
What this paper finds — and why it matters
Using unpublished BLS microdata behind the CPI for 1995-1997 – about 350 categories covering roughly 70% of consumer spending – Bils and Klenow find that individual retail prices change much more often than earlier studies of narrower samples had suggested: the weighted median category’s price changes about every 4.3 months, and the weighted mean monthly frequency of a price change is 26.1%. Price flexibility varies enormously across goods and is systematically related to market structure: categories with higher four-firm concentration ratios and higher wholesale markups change price less often, while categories with a higher rate of product turnover (new models/varieties replacing old ones) change price more often, and raw, less-processed goods (gasoline, fresh produce) reprice far more often than processed goods and services. The paper then asks whether the standard Calvo/Taylor time-dependent sticky-price models – calibrated to these same measured frequencies – can match how inflation for a given good actually behaves: in those models greater price stickiness should make a good’s inflation more persistent and its innovations less volatile. Comparing model-implied persistence and volatility to the actual behavior of inflation for 123 of the matched goods over 1959-2000 and 1995-2000, the authors find inflation is far more volatile and far less persistent than the calibrated Calvo/Taylor models predict, and the mismatch is worst for goods with the least frequent price changes – a pattern that survives adding plausible idiosyncratic productivity shocks and is not an artifact of seasonal adjustment. The authors conclude that popular time-dependent sticky-price models, taken at face value with empirically measured frequencies, do not capture the actual dynamics of sectoral inflation.
Summary of a classic paper, AI-assisted and human-reviewed. See the linked original for the authoritative claims and full conditions.
Questions & answers
Q1. What data underlie the paper’s claim that prices change much more often than earlier studies found?
The paper uses unpublished BLS “Commodities and Services Substitution Rate” tables for 1995-1997, covering 350 Entry Level Items (ELIs) that together make up 68.9% of consumer spending in the 1995 Consumer Expenditure Survey. These tables record, for each ELI, what share of monthly (or, outside the five largest metro areas, bimonthly) price quotes differed from the previous quote for the same item at the same outlet; the authors back out an implied constant monthly hazard of a price change for each ELI. The weighted median monthly frequency is 20.9%, implying a median time between price changes of 4.3 months, and the weighted mean is 26.1%. This is far more frequent adjustment than survey-based or narrower-sample studies had found: Blinder et al.’s (1998) survey of 200 firms found a median adjustment of about once a year, and narrower consumer-price studies (Cecchetti’s 1986 magazine prices, Kashyap’s 1995 mail-order catalog goods) implied adjustment roughly every 11-15 months, versus 4.3 months for the median BLS category.
Q2. What predicts whether a good’s price is flexible or sticky?
Prices change less often in more concentrated, higher-markup industries and more often where products turn over quickly, and raw/unprocessed goods reprice far more than processed goods and services. Regressing the frequency of price change on the industry’s four-firm concentration ratio, wholesale markup, and rate of “noncomparable” item substitutions (a proxy for product turnover), the authors find all three enter with the expected sign and are statistically significant; the concentration-ratio coefficient alone implies that moving from the 23% concentration of pet food to the 99% concentration of cigarettes is associated with more than a 20-percentage-point lower monthly frequency of price change. Once a raw-goods dummy is added, the concentration and markup coefficients become unimportant but the product-turnover variable remains robust, with a 1% higher rate of noncomparable substitutions associated with roughly a 2-2.9% higher frequency of price change. Raw goods (gasoline, meat, fresh produce) reprice 54% of months on average versus 21% for processed goods and services.
Q3. How does the paper build a general-equilibrium model to generate predictions for sticky- versus flexible-price-sector inflation?
The authors adapt the two-sector monopolistic-competition, money-in-the-utility-function model of Chari, Kehoe, and McGrattan (2000), splitting consumption goods into a flexible-price sector (prices preset for 2 periods, the 90th percentile of the empirical frequency distribution) and a sticky-price sector (prices preset for 15 periods, the 10th percentile), with staggered, Taylor-style price setting within each sector. Firms are monopolistically competitive with a 10% desired markup, production is linear in labor with random-walk productivity, and money growth follows an AR(1) process calibrated to U.S. M1 growth. In response to a permanent 1% money-supply shock, the model’s sticky sector shows a more persistent and less volatile inflation response than the flexible sector, because successive cohorts of price-setters adjust gradually; the same qualitative pattern holds for a permanent technology shock.
Q4. What does the paper find when it tests this model’s predictions against actual sectoral inflation data?
Matching the 350 BLS categories to 123 NIPA price series covering 63.3% of 1995 consumer spending, the paper finds that inflation for the median consumer good is far less persistent, and typically more volatile, than the calibrated Calvo/Taylor model predicts. Over 1995-2000, the average AR(1) serial correlation of monthly inflation across the 123 goods is close to zero (-0.05, s.e. 0.02), and goods with more frequent price changes actually show more (not less) persistence (correlation of 0.26 between frequency and serial correlation), the opposite of the sticky-price-model prediction; consistent with the models, however, goods with more frequent changes do show more volatile inflation innovations (correlation 0.68). Formally comparing simulated model statistics to the data (Table 7/8 comparisons), for a hypothetical sticky-price good (monthly frequency 1/15) the Calvo/Taylor model predicts inflation persistence around 0.91, while the data over 1959-2000 show persistence of only about 0.28, and -0.11 over 1995-2000 – a large and, for less-frequently-repriced goods, essentially universal mismatch.
Q5. Could the mismatch be explained away by adding idiosyncratic shocks or more persistent marginal-cost shocks?
No – the authors show the gap is robust to adding realistic idiosyncratic productivity shocks calibrated to industry TFP data, and that reconciling the model’s persistence with the data by assuming less-persistent idiosyncratic cost shocks would require implausibly large shock variances. Adding idiosyncratic TFP shocks (calibrated to 459 manufacturing industries, with autocorrelation in levels of 0.98) raises inflation volatility in the model by about 50% but barely changes predicted persistence (from 0.91 to 0.90 for the sticky good). To make the sticky-price model’s persistence match the data by instead assuming serially-uncorrelated-in-levels idiosyncratic shocks, the authors calculate that the implied monthly standard deviation of idiosyncratic productivity in the sticky sector would have to be an implausible 59%.
Q6. Did the authors test whether seasonal price patterns were driving the transience result?
No – removing monthly seasonal dummies from the inflation series left the persistence and volatility findings essentially unchanged, so synchronized seasonal pricing (such as regular seasonal sales) is not what produces the low measured persistence of goods’ inflation.
Q7. What alternative class of models does the paper suggest might better fit these facts?
The authors suggest the facts may be easier to reconcile with state-dependent pricing models, in which the frequency of price adjustment responds endogenously to the size of shocks, since such models (citing Willis, 2000) can generate synchronized price responses to large sectoral shocks that produce large, transient inflation innovations rather than the smooth, persistent inflation that time-dependent Calvo/Taylor models imply.
Key terms in this paper
Definitions below follow the paper's own usage.
- Entry Level Item (ELI)
- the BLS's basic pricing category (350 of them used here, covering 68.9% of 1995 consumer spending); the unit at which the unpublished BLS Commodities and Services Substitution Rate tables report the share of monthly price quotes that changed.
- Frequency of price change
- the estimated constant monthly probability that an item's price differs from its value the previous month, backed out from the mixed monthly/bimonthly BLS substitution-rate data; the paper's central measurement, ranging from 1.2% to 79% across the 350 ELIs, with a weighted mean of 26.1% and weighted median of 20.9%.
- Noncomparable substitution
- a BLS-classified item substitution (when a priced item is discontinued at an outlet and a field agent selects a replacement) that is judged not comparable in quality to the discontinued item and so enters the CPI with a quality adjustment; its rate is the paper's proxy for product turnover and is the most robust predictor of price-change frequency across the regressions in Table 4.
- Calvo/Taylor time-dependent pricing
- the workhorse sticky-price framework in which a firm's probability (Calvo) or the duration (Taylor) of holding a price fixed is exogenous and unrelated to the size of shocks; the paper's central empirical test is whether, calibrated to the BLS-measured frequencies, these models' predicted sectoral inflation persistence and volatility match the data -- they find the models imply far more persistence and far less volatility than observed, especially for infrequently-repriced goods.
- Raw versus processed goods
- the paper's distinction between goods close to a primary input (gasoline, fresh produce, meats) with volatile, undiversified costs, and goods/services with more value added beyond the primary input; raw goods reprice roughly two and a half times as often (54% of months) as processed goods and services (21%).