Abstract

We study the impact of targeted price controls on supermarket products in Argentina. Using daily prices for controlled and noncontrolled goods, we examine the effects on inflation, product availability, entry and exit, and price dispersion. First, price controls have only a small and temporary effect on inflation that reverses soon after the controls are lifted. Second, contrary to common belief, controlled goods are consistently available for sale. Third, firms respond by introducing new varieties at higher prices. Overall, our results show that targeted price controls are just as ineffective as more traditional policies of price controls in reducing aggregate inflation.

I. Introduction

MANY governments consider the use of price controls when inflation rises. In some cases, these price controls are imposed across the board, affecting all goods and causing widespread shortages.1 Hoping to avoid these negative effects, governments often implement a more subtle form of targeted price controls on a limited number of selected goods.2 Traditionally, these controls have been limited to highly regulated sectors, such as pharmaceuticals, utilities, and gas prices.3 More recently, the availability of online data, mobile phone apps, and electronic records has dramatically increased the ability of governments to implement, monitor, and enforce targeted price controls in all kinds of consumer goods. In particular, developing countries such as Argentina, Ecuador, Israel, and Panama have recently had some form of targeted price controls for food and grocery products.

Despite the increased interest in the use of targeted price controls, there is little empirical research documenting their economic effects. Are these controls binding? Do they affect prices and inflation of related goods? Can they avoid shortages associated with more generalized controls? What strategies do firms employ to deal with potentially lower profits and better enforcement?

In this paper, we answer some of these questions by studying the rich and volatile experience with price controls in Argentina from 2007 to 2015. These were targeted controls, affecting a selected set of consumer products identified at the bar code level. To help with the visibility and enforcement programs, the government required retailers to display labels identifying individual goods as being under a “government agreement” in both offline and online stores. We used Web-scraping methods to identify these price-controlled goods and track their prices on a daily basis from October 2007 to May 2015. Our data include more than 50,000 products sold by one of the largest supermarkets, including approximately 1,400 goods that were under price control at some point during this period. These controls were focused on goods that have significant weights in the CPI basket and for varieties sold by leading brands.

With this high-frequency panel of controlled and noncontrolled goods, we examine the before-and-after impact of price controls on inflation, product availability, and price dispersion. Our main results are summarized as follows.

First, the impact of targeted price controls on aggregate inflation is small and temporary. At the microlevel, controlled goods are sold at government-agreed prices, which are on average 3.3% lower than before the control, but these are compensated by similar price increases soon after the controls are removed. At the macrolevel, the inflation rate of controlled goods fluctuates between periods of marginally lower and higher inflation relative to other goods, with no significant impact on the aggregate inflation rate.

Second, contrary to common belief, we find that controlled goods are seldom discontinued and that their availability is similar to that of noncontrolled goods. They have a higher probability of going temporarily out of stock, but stockouts are short-lived and goods are only occasionally discontinued. This suggests that the government's ability to enforce both prices and stocks was effectively enhanced by new monitoring technologies, including an official mobile app that allowed consumers to scan product bar codes and send online complaints to the enforcement agency in cases of price discrepancies or stockouts.

Third, given that price controls are binding in prices and availability, we study how firms might offset lower profit margins. Consistent with the predictions of a standard vertical differentiation model in the presence of price controls, we find evidence that firms expanded their product line with new varieties at higher prices. These new varieties tend to be associated with additional characteristics that suggest a quality improvement (e.g., milk with “extra calcium”), implicitly reducing the perceived quality of older varieties under price controls. This firm-level strategy raises average prices and price dispersion within good categories with controlled goods and potentially introduces a welfare cost to consumers by obfuscating choices and increasing search frictions.

Table 1.
Time Line of Price Controls
Isolated ControlsTotal FreezeLook to CareProtected Prices
Stage 1Stage 2Stage 3Stage 4
Period Oct. 2007 to Feb. 2013 Feb. 2013 to May 2013 June 2013 to Nov. 2013 Jan. 2014 to May 2015 
Public Information No No Yes Yes 
Same Goods All Retailers No No No Yes 
Target Number of Products — All 500 100 
Isolated ControlsTotal FreezeLook to CareProtected Prices
Stage 1Stage 2Stage 3Stage 4
Period Oct. 2007 to Feb. 2013 Feb. 2013 to May 2013 June 2013 to Nov. 2013 Jan. 2014 to May 2015 
Public Information No No Yes Yes 
Same Goods All Retailers No No No Yes 
Target Number of Products — All 500 100 

Stage 2 is excluded because the government aimed to freeze all food products and did not officially disclose the identifiers of controlled products. Our scraping algorithm identified a limited number of controlled goods (section III). Our data end in May 2015, but the Protected Prices program continued after that. Details about the programs are in section II.

Overall, our results suggest that targeted price controls are just as ineffective as more traditional forms of price controls in terms of reducing aggregate inflation, despite the advances in targeting, monitoring, and enforcement technologies. These findings complement a large macroeconomic literature that focuses on U.S. price and wage controls during Richard Nixon's presidency in the 1970s.4 Our work is also related to papers that study price controls from a micro or industrial organization perspective. Leffler (1982) provides a model where firms decrease quality until shortages from binding maximum prices are eliminated. Besanko, Donnenfeld, and White (1987, 1988) find that maximum price regulation can counteract the quality distortion in a monopoly price setting, and that firms may lower quality for lower willingness-to-pay consumers. Raymon (1983) argues that binding price ceilings can decrease quality and consumer welfare in competitive markets, while Kyle (2007) finds that price controls in one market affect entry strategies and the introduction of new products in other markets, consistent with our findings.

The remainder of the paper proceeds as follows. Section II describes price controls in Argentina from 2007 to 2015. Section III describes the Web-scraping technology and the micro data set. Sections IV and V discuss the impact of price controls on inflation and product availability, respectively. Section VI presents evidence of firms' strategic behavior in the presence of controls. Section VII shows robustness specifications. Section VIII concludes.

II. Price Controls in Argentina

Argentina has a long history of price controls. In 1939, Congress passed a law to prevent stockouts during World War II. And although price controls were originally conceived as a temporary means to lower inflation, subsequent governments have continued to rely on various forms of such policy.5

We study the 2007–2015 period, when the government experimented with various types of targeted price controls in supermarket products. These price controls were meant to curb inflation, which rose from 10% in 2006 to over 35% in 2016 (according to unofficial estimates).6 The programs focused on food and beverages, which constitute nearly 40% of the CPI basket. Despite the fact that they did not appear to have much of an impact on the inflation rate, these programs were popular with voters. According to recent consulting and media surveys, 60% of Argentines supported price-control policies, and 25% of consumers bought price-controlled goods, which accounted for up to 20% of retailers' revenues in supermarkets.7

This period of price controls can be divided into four stages, as summarized in table 1. Stages 1, 3, and 4 are examples of targeted price controls, and stage 2 was a temporary freeze of all goods sold by large supermarkets. In all cases, the control price applies to the retail price inclusive of VAT taxes.8

The first stage lasted from 2007 to 2013 and was characterized by confidential ad hoc price agreements with major supermarkets, which had to freeze prices temporarily for selected goods. No official press releases or announcements disclosed the specific products being controlled, but some retailers showed a “government agreement” label next to the product. We use this label to identify controlled products through scraping technologies. News articles throughout the period reported that this policy resulted in major shortages (Raszewski, 2007), but we find no evidence of consistent shortages in our data, as we discuss later.

The second (and only nontargeted) stage started in February 2013, when unofficial inflation reached 25%. The government announced that it had reached a new agreement with the largest retailers in Argentina to freeze prices of all products for sixty days and later extended it for another sixty days until May 31 (Raszewski, 2013). As before, no official documentation was released. The indiscriminate attempt to control all prices, together with an increased duration of controls, might explain why the number of products available to the public started to decrease around this time, as discussed in section V. Prices remained stable for a few weeks but soon started to rise again, prompting the government to adjust its price control policies.

Figure 1.

Intensity Index

Retailers' narrow categories under price controls. Vertical notes denote the four main events in the time line as described in the main text.

Figure 1.

Intensity Index

Retailers' narrow categories under price controls. Vertical notes denote the four main events in the time line as described in the main text.

A third stage started in June 2013, when the government lifted the controls on all but 500 products. This time, the government published the names and government-agreed prices for all controlled products, which included food (fresh produce and packaged), beverages, cleaning, and health and personal care items. It first targeted major retailers in Buenos Aires and then expanded throughout the country. Because not all retailers sold the same brands or categories, each supermarket had its own list of 500 product varieties whose prices had to stay constant for six months. This price agreement was formalized under the name “Mirar para Cuidar” (Look to Care). The government increased the program enforcement as well as its advertising in public media. La Campora, a political organization with close ties to the president, developed a Web and mobile phone app that allowed militants to help monitor and enforce the price controls. Several store locations were temporarily closed or fined due to shortages.9

Figure 1 shows how the government increased the intensity of the price controls starting in 2013, relative to the first stage of targeted controls. We measure the intensity by counting the number of distinct retailer categories (URLs) that had controlled goods during each month.10

In December 2013, amid significant changes in the cabinet, the government announced a new stage of price controls, Precios Cuidados (Protected Prices).11 Launched in January 2014, it drastically reduced the product list to 100 different goods (194 varieties in total) that were by then common among all major retailers. The duration of the price controls also increased significantly: the median time a product remained under price controls increased from 70 days in the first stage to 183 days in the third stage. In an attempt to facilitate the diffusion of price lists, the new program started with fewer categories and varieties and steadily increased over time. Protected Prices also required producers to inform the government of new product introductions that resembled those under control. This clause was added amid criticism that in previous controls, firms launched similar products or varieties to circumvent maximum prices.12 We discuss evidence of that strategy in section VI.

The government also increased the firms' costs of violating the agreements and implemented tighter and more sophisticated monitoring strategies.13 The government developed a website with all product lists and prices and made the information accessible with a mobile phone app that allowed consumers to scan product bar codes and report stockouts or incorrect prices. Militant groups close to the government posted pictures of CEOs and owners of supermarket chains in the streets and encouraged people to help monitor prices. These strategies were extensively advertised in public media,14 including radio, television, newspapers, and official press releases.15

Table 2.
Summary Statistics
All GoodsNoncontrolledControlled Goods
(i) Time period October 2007 to May 2015 October 2007 to May 2015 October 2007 to May 2015 
(ii) Observations (with price) 15,796,787 15,139,656 657,131 
(iii) Distinct goods 51,779 50,319 1,460 
(iv) Distinct brands 3,518 3,466 438 
(vi) Retailer's categoriesa 964 963 302 
(vi) CPI categories 75 75 54 
(vii) Total CPI weight 44.6 44.6 39.7 
(viii) Average CPI weight per category 0.859 0.802 0.923 
(ix) Average CPI weight per product 0.860 0.853 1.094 
(x) Average CPI weight per productb 0.001 0.001 0.034 
(xi) Median control time — — 75 days 
(xii) Median control eventsc — — 2 times 
(xiii) Percent of time under controld — — 23% 
(xiv) First control at higher pricee — — 6% 
(xv) First control at lower price — — 32% 
(xvi) First control at same price — — 51% 
(xvii) Average price change at controlf — — -3.3% 
All GoodsNoncontrolledControlled Goods
(i) Time period October 2007 to May 2015 October 2007 to May 2015 October 2007 to May 2015 
(ii) Observations (with price) 15,796,787 15,139,656 657,131 
(iii) Distinct goods 51,779 50,319 1,460 
(iv) Distinct brands 3,518 3,466 438 
(vi) Retailer's categoriesa 964 963 302 
(vi) CPI categories 75 75 54 
(vii) Total CPI weight 44.6 44.6 39.7 
(viii) Average CPI weight per category 0.859 0.802 0.923 
(ix) Average CPI weight per product 0.860 0.853 1.094 
(x) Average CPI weight per productb 0.001 0.001 0.034 
(xi) Median control time — — 75 days 
(xii) Median control eventsc — — 2 times 
(xiii) Percent of time under controld — — 23% 
(xiv) First control at higher pricee — — 6% 
(xv) First control at lower price — — 32% 
(xvi) First control at same price — — 51% 
(xvii) Average price change at controlf — — -3.3% 

aWebsite retail categories (e.g., dairy), which are broader than URL-based retail subcategories (e.g., yogurt). bWeighted by number of products in each category (e.g., if a category weights 3 and there are 10 products, then each product's weight is 0.3), then averaged across all goods. cNumber of (nonconsecutive) times a product received price controls, and then median across controlled goods. dCalculated using nonmissing observations (in stock for sale). fFraction of controlled goods whose first control was set at a higher price, relative to its last available price without controls. Similarly for “First control at lower price” and “First control at same price.” The remaining fractions are new items and have no price change available. eBased on the average price ten days before and after the first control.

As of the time of writing, Protected Prices remains active, with new products and maximum prices announced quarterly.16 Price controls remain popular with consumers in Argentina, which is consistent with the experience in the 1970s in the United States (Nixon, 1978; Blinder, 1979) and Sweden (Jonung, 1990).

III. Scraped Online Prices

We use online prices from thousands of products sold online each day from 2007 to 2015 by one of the largest retailers in Argentina in terms of market share.17 The data were scraped off the Internet by the Billion Prices Project, an academic initiative at Harvard and MIT that collects online prices from hundreds of retailers around the world (Cavallo & Rigobon, 2016).

The scraping software is designed to search the HTML public code of a retailer's website and automatically store the pricing data of all goods on a daily basis. The retailer assigns a unique ID to each product sold online. In the days when the scraping fails (due to software failures or web page maintenance) prices are assumed to remain constant until the goods are back online. Goods that do not reappear on the website are considered to be discontinued.

We identified price-controlled goods in two ways. First, from 2007 to 2015, the scraping algorithm read a special HTML (ID-specific) tag next to each controlled good sold online. This method accounts for about 75% of the controlled goods in our database. Second, after the government started publishing lists of controlled goods in 2013, we manually identified each of these goods in our database.18

A. Data description

Table 2 provides summary statistics of the data coverage. We have daily prices for more than 50,000 distinct products from 2007 to 2015 and a yearly average of about 14,000 distinct goods. The supermarket scraped data cover categories such as food, beverages, electronics, household appliances, kitchen utensils, and health and personal care items, which collectively account for about 45% of the CPI weights. (See Aparicio & Rigobon, 2020, for additional details on Web scraping and decision tree classifiers to categorize product-level data. See Gorodnichenko, Sheremirov, & Talavera, 2018, and Cavallo, 2018, for related work that documents price-setting behaviors in the online channel.)

We label products as controlled if they were affected by a price control for least one day during the scraping period.19 This results in 1,460 controlled goods, which is about 3% of the total products in the database. Although a relatively small set, these goods have a significantly higher weight in the CPI basket, as seen in table 2.20

On average, goods were controlled for six months, with a median of two and a half months. About 25% of the controlled goods had price controls lasting more than seven months, and if a product price was controlled more than once, there were usually no gaps in between. Controlled goods were under price controls about 23% of their time available online. And a stable subset of items consistently remained under price controls throughout the scraping period.21 Price controls were generally imposed at the existing price level, but in about a third of the cases, the new price was set lower. On average, the price change was -3.3%.

IV. Prices and Inflation

To study the impact of price controls on inflation, we construct a simple aggregate price index as a weighted arithmetic average of category indices. Each category index is a Jevons geometric average of all products sold online. If an item is out of stock, we assume constant prices. And if an item is discontinued, then it no longer has an impact on the index. Products that are out-of-stock are momentarily unavailable for online purchase, while discontinued products were no longer offered online until the end of our scraping period. Price changes are weighted using Argentina's National Statistics Office (INDEC) official weights by CPI category. (See Cavallo, 2013, and Aparicio & Bertolotto, 2020, for evidence that online price indices closely track and forecast official CPIs.)

Figure 2 shows the price indices and the annual and monthly inflation rates for three samples: all goods, controlled goods, and noncontrolled goods.
Figure 2.

Price Index and Inflation Rates for Different Samples

The price index is calculated using official weights by CPI categories and unweighted geometric averages of price changes for subcategories without official weights. The annual and monthly inflation rates are computed using a thirty-day moving average of the price index.

Figure 2.

Price Index and Inflation Rates for Different Samples

The price index is calculated using official weights by CPI categories and unweighted geometric averages of price changes for subcategories without official weights. The annual and monthly inflation rates are computed using a thirty-day moving average of the price index.

The impact of price controls on aggregate inflation is small and temporary. From 2007 to 2015, all price indices had recorded about 400% accumulated inflation. There are periods when the inflation rate of controlled goods appears significantly different from that of noncontrolled varieties, as shown in the volatility of the monthly inflation rate in panel c. These periods are associated with weak or strong periods of price agreements, but the difference was never large enough to have a significant impact on the aggregate inflation rate for all items, as shown in panel b. The lack of expenditure weights within categories means that we could be underestimating the effects if the government targeted leading brands, but the impact was nevertheless temporary.

Table 3.
Effects on Product Monthly Inflation
Product Inflation
(1)(2)(3)(4)
Product controlled −0.837*** −0.793***   
 (0.073) (0.073)   
Product with controlled competitor  0.165***   
  (0.024)   
Control lifted   4.904*** 1.802*** 
   (0.334) (0.452) 
Extended control lifted    4.052*** 
    (0.608) 
Category FE YES YES YES YES 
Time FE YES YES YES YES 
Observations 1,546,383 1,546,383 1,546,383 1,546,383 
Product Inflation
(1)(2)(3)(4)
Product controlled −0.837*** −0.793***   
 (0.073) (0.073)   
Product with controlled competitor  0.165***   
  (0.024)   
Control lifted   4.904*** 1.802*** 
   (0.334) (0.452) 
Extended control lifted    4.052*** 
    (0.608) 
Category FE YES YES YES YES 
Time FE YES YES YES YES 
Observations 1,546,383 1,546,383 1,546,383 1,546,383 

Observations are aggregated to the product and month level. Dependent variable is the percentage change in the product-specific monthly average price. Estimates remain quantitatively similar if covariates are added sequentially or together. Robust standard errors in parentheses: ***p<0.01, **p<0.05, and *p<0.1.

Periods of low inflation in controlled goods were quickly followed by higher inflation for those same goods. In particular, controlled goods had lower inflation until 2009 and then much higher inflation in 2010 and 2011.22

Table 3 shows the micro impact of price controls using product-specific regressions. We calculate the monthly average price for each product and run regressions of the form
πti,j=a+βDti+γt+μj+eti,
(1)
where πti,j is the percent change in monthly price of product i from category j at time t; Dti is a price control indicator; and γt and μj are time- and category- fixed effects, respectively.

In column 1, we define an indicator that takes value 1 if the product experienced a price control during month t. The estimate suggests that products have 0.84% lower inflation during the month of price controls. In column 2, we include an indicator if a noncontrolled product has a competitor under price controls (defined as in the same narrow category). The point estimate suggests that noncontrolled products experience about 0.17% higher inflation when a competitor is under price controls, suggesting that retailers are not able to fully compensate by increasing prices of noncontrolled related goods.

Column 3 shows that the effect on inflation is temporary, with prices rising as soon as the price control is lifted. In this case, we use an indicator variable that takes a value of 1 if the product experienced a price control at month t-1 but not at t. The estimated coefficient suggests that monthly inflation increases by an additional 4.9 percentage points as soon as price controls are lifted.

We next explore whether the effect varies with the duration of the control. We include an indicator that takes the value 1 if the product experienced price controls during at least three consecutive months (but not in month t). Controls that lasted less than three months cause an increase of 1.8 percentage points, while those that last more than three months lead to an increase of approximately 5.9 percentage points (adding both coefficients).

Overall, these results suggest that price controls have a temporary effect on the inflation of controlled products. These regressions include results from goods that may have been under price controls multiple times, allowing the firms to act strategically with their pricing. In the next section, we explore high-frequency pricing reactions using an event study that focuses exclusively on the first time each product was affected by a price control, when the impact is presumably higher.

A. Effect on First-Time Controls

We now use high-frequency price data on the first time each controlled good experienced a price control. In order to determine whether firms increased the price of related goods after price controls were imposed, we split the sample of noncontrolled goods into “related” and “unrelated” products. Related noncontrolled goods are sold in subcategories that have goods under price controls at the same time—for example, cereals. Unrelated noncontrolled goods are sold in subcategories that had no price controls at all.

The related sample is constructed as follows. Each time an item is controlled for the first time, we randomly select another product from the same subcategory—for example, (a) it is in stock that day, (b) it is not controlled during the scraping period, or (c) it has not already been selected as related to another good (i.e., draws without replacement). The unrelated sample is constructed in a similar way but from noncontrolled categories. The controlled, related, and unrelated samples comprise 1,460, 1,321, and 1,400 distinct items, respectively.

For each good, we keep prices for ninety days before and after it received the first price control.23 This method produces a balanced panel for each sample (except for some censored observations in some price spells for discontinued goods). We then compute the thirty-day rolling inflation at the good-level and average it across goods each day. This generates an approximation of the average monthly inflation, shown in figure 3 for a ninety-day window around the day the control is first introduced.
Figure 3.

High-Frequency Event Study in a Ninety-Day Window

Monthly inflation rate is computed separately for controlled products and for noncontrolled related and unrelated products. The price index uses price data on ninety days before and after each product received the first price control.

Figure 3.

High-Frequency Event Study in a Ninety-Day Window

Monthly inflation rate is computed separately for controlled products and for noncontrolled related and unrelated products. The price index uses price data on ninety days before and after each product received the first price control.

This figure highlights three findings. First, consistent with the previous discussion, there were temporary effects on controlled-goods prices. The monthly inflation rate falls to −5% immediately after the control is imposed, and thirty days later, the monthly inflation is close to 0% (indicating that prices remain fixed). However, after two months, the inflation rate jumps to +5%, compensating for previous declines.

Second, noncontrolled goods in related categories do not have higher inflation rates after the control is introduced. Their behavior is similar to noncontrolled goods in unrelated categories, suggesting that at least immediately after the first event, firms were not compensating for the controls by increasing prices of related goods. In some cases, the government explicitly monitored the behavior of related goods, which likely limited firms' ability to compensate through the prices of other existing varieties.

Third, related goods do have higher inflation before the controls are imposed. Their prices start to rise faster sixty days before the control, and the inflation difference peaks fifteen days before the control is imposed. Goods that will later be under a price control and those in unrelated categories do not experience similar price increases. One possible explanation is that firms increased prices of some varieties to make potential candidates for a price control appear relatively cheap. Overpricing varieties that were less likely to be under a price control could be a strategy to negotiate a higher price ceiling on the cheap variety.24

In summary, first-time price controls on specific goods did not have a downward effect on the inflation rate of related goods in the same categories. In section VI we explore whether targeted price controls affected the price of new varieties introduced after a price control takes place.

V. Product Availability

Can better monitoring tools for targeted controls prevent the shortages typically associated with more extended types of price controls? In order to answer this, we compute a measure of product availability, defined as the number of items available for sale online on a given day.25 Figure 4a shows that the retailer sold over 13,000 products per day, 700 of them controlled-goods. The flat line between late 2009 and early 2010 is due to a partial scraping failure in our algorithm. Scraping failures are otherwise only occasional and do not affect the data.

The availability of controlled goods was surprisingly stable over time, at around 700 items per day. Although newspapers claimed that price controls produced major stockouts, we found no such evidence in our data. We also simulated online purchases on several occasions and found no shipping delays or limits on the number of units that could be purchased.26

The only major drop in availability occurred when the government imposed a total price freeze in stage 2 (shaded region). About 100 goods were discontinued during that period, and another 200 disappeared when controls ended. Once the programs became targeted again, in stage 3, the availability of controlled goods stabilized, though at a much lower level. It is possible that as intensity and duration of price controls increased, retailers decided to discontinue products that were previously controlled.27

Figure 4.

Product Availability

Product availability index is defined as the number of distinct products available for sale online each day.

Figure 4.

Product Availability

Product availability index is defined as the number of distinct products available for sale online each day.

A. Temporary Stockouts

Even if the government can prevent retailers from discontinuing goods, we might expect controlled goods to experience frequent stockouts. In this section, we use survival analysis to study the risk of stockouts across samples.

The onset of risk, or t0, is defined as the day each good received its first price control during the scraping period from 2007 to 2015. The end date, or failure event, is the day of the first stockout after the control is imposed. If the scraping package fails, no price observations are recorded for that date. We control for these cases and for right-censored observations (i.e., controlled goods that did not go out of stock by the end of the scraping period).

Figure 5a shows a histogram of the number of days until the first stockout, that is, we compute the number of days between t0 and the failure event for each good. We find that controlled goods do experience a relatively faster stockout: one and a half months after the first price control, more than 50% of the goods have gone out of stock compared to 40% in the related noncontrolled varieties. Vertical lines depict average days for each sample.

We also estimate the survival function S(t), defined as the survival probability (or in-stock probability) past time t, that is, the probability of failing after t, for both controlled and related goods. We use the nonparametric algorithm from Kaplan and Meier (1958):
S^(t)=j|tjtnj-djnn,
where nj is the number of goods at time tj and dj is the number of stockout events at time tj, and where the product is computed over all observed failure times until time t.

Figure 5b shows similar estimated survival function for both controlled and related goods.28 We find that the probability of being in stock is about 15% higher for (related) noncontrolled goods a month after controls were imposed. The magnitude of this probability seems small, particularly when we consider that the average duration of a stockout (defined as the number of days out of stock after each product's first control) is only 3 days for controlled good and 2.6 days for related goods

VI. New Varieties and Price Dispersion

The previous sections show that targeted controls in Argentina do not significantly affect the aggregate inflation rate, though they do force some firms to sell goods at lower prices and keep them in stock most of the time. So how do firms cope with price controls?29

Figure 5.

Stockout Behavior

Histogram computed for less than six months for better visualization. Median days until first stockout in vertical lines. Kaplan-Meier survival function is computed for all months, but axis is also restricted to six months.

Figure 5.

Stockout Behavior

Histogram computed for less than six months for better visualization. Median days until first stockout in vertical lines. Kaplan-Meier survival function is computed for all months, but axis is also restricted to six months.

For targeted price controls, an effective strategy might be to introduce new varieties at higher prices while implicitly suggesting to consumers that the older varieties have lower quality.30 For example, in June 2013, Sancor's Dulce de Leche, 1 kg was controlled; a few weeks later Sancor introduced Dulce de Leche, Countryside Style, 400 grams at a size-adjusted 33% price increase. Examples like these are not easy to find, however, because firms obfuscate the comparisons with complicated product descriptions or package size changes. In the appendix, we show that this firm-level strategy is consistent with a vertical differentiation model in the presence of targeted price controls.

We document how the number of noncontrolled varieties increases by running the following regressions at the category-URL and month level:
Varietiestj=a+βDtj+γt+μj+eti,
(2)
where Varietiestj is the number of noncontrolled varieties (in logs) in category-URL j at month t; γt and μj are time and category fixed effects, respectively; and Dtj is an indicator that takes value 1 when category-URL j has at least one product under price controls at month t. The main estimate in column 2 of table 4 indicates that narrow categories subject to price controls experience a 33.7% increase in new noncontrolled varieties.31

Traditional matched-model price indices, such as those used in section IV, are unable to capture the price impact of new varieties. The reason is that they are based on the price changes of goods that are present in two time periods, and therefore the price levels of new products are not included in the index computations.

Table 4.
Effects on Noncontrolled Varieties and Price Dispersion
NC VarietiesPrice Dispersion
(1)(2)(3)(4)
1{Controlledthismonth} 0.418*** 0.337*** 0.052*** 0.083*** 
 (0.0866) (0.0708) (0.018) (0.019) 
πt    −0.002 
    (0.003) 
Constant 2.497*** 1.404*** 0.466*** 0.750*** 
 (0.0313) (0.0154) (0.010) (0.004) 
Category FE NO YES NO YES 
Time FE NO YES NO YES 
Observations 65,276 65,021 68,077 67,771 
NC VarietiesPrice Dispersion
(1)(2)(3)(4)
1{Controlledthismonth} 0.418*** 0.337*** 0.052*** 0.083*** 
 (0.0866) (0.0708) (0.018) (0.019) 
πt    −0.002 
    (0.003) 
Constant 2.497*** 1.404*** 0.466*** 0.750*** 
 (0.0313) (0.0154) (0.010) (0.004) 
Category FE NO YES NO YES 
Time FE NO YES NO YES 
Observations 65,276 65,021 68,077 67,771 

Observations are aggregated to the category-URL and month level. In columns 1 and 2, the dependent variable is the log number of distinct (nonmissing) noncontrolled goods. In columns 3 and 4, the dependent variable is the coefficient of variation. We let 1{Controlledthismonth} take 1 when a price ceiling affects any good on a certain URL-month. π is the online-measured monthly inflation rate. Coefficient remains similar if we control for inflation volatility and exchange rate depreciation. Categories are CPI subcategories. Standard errors clustered at the URL level in parentheses: ***p<0.01, **p<0.05, and *p<0.1.

A simple way of detecting the impact of new varieties is by constructing average price indices, which measure the average price level for all varieties sold each day in a narrowly defined category.32 We first compute average-price indices for both controlled and noncontrolled goods in the same subcategories and then build aggregate price indices using official CPI category weights. Our data are well suited for this analysis because the Web-scraping algorithm adds goods to our sample on the first day they appear on the store.

As figure 6a shows, the inflation rate for noncontrolled goods is higher once we account for the price levels of new varieties at the time of introduction. The average-price index has more inflation than the corresponding matched-model (chained) index that uses only price changes and significantly more inflation than the average-price index for controlled goods.
Figure 6.

Price Index and Price Dispersion

(a) “Controlled” and “Non-Controlled” are average price indices as described in the text; “Non-Controlled (Chained)” is a standard chained or matched-model price index. (b) Price dispersion is measured in narrowly defined categories before and after products receive the first price control.

Figure 6.

Price Index and Price Dispersion

(a) “Controlled” and “Non-Controlled” are average price indices as described in the text; “Non-Controlled (Chained)” is a standard chained or matched-model price index. (b) Price dispersion is measured in narrowly defined categories before and after products receive the first price control.

Table 5.
Inflation, Stockout, and New Varieties by Stage
Isolated ControlsGeneral FreezeLook to CareProtected Prices
Stage 1Stage 2Stage 3Stage 4All Stages
A - Monthly inflation:a      
Products controlled −0.78 2.37 −1.96 −0.43 −0.84 
B - Probability of stockout:b      
30 days after 0.27 0.23 0.20 0.42 0.28 
C - New varieties:c      
Products controlled 0.32 −0.28 0.49 0.51 0.34 
Isolated ControlsGeneral FreezeLook to CareProtected Prices
Stage 1Stage 2Stage 3Stage 4All Stages
A - Monthly inflation:a      
Products controlled −0.78 2.37 −1.96 −0.43 −0.84 
B - Probability of stockout:b      
30 days after 0.27 0.23 0.20 0.42 0.28 
C - New varieties:c      
Products controlled 0.32 −0.28 0.49 0.51 0.34 

aCoefficient from the regression of product-specific monthly inflation rate (in percentage) on an indicator that takes 1 if the product received a price control during month t. Sample restricted to each stage period. The regression follows equation (1), which includes category and time fixed effects. All coefficients are statistically significant at the 1% level. Complete estimates for the entire period are shown in table 3. bKaplan-Meier inverse probability of survival. This reproduces the survival analysis in of figure 5b. cCoefficient from the regression of category-URL noncontrolled varieties (in logs) on an indicator that takes 1 if the product received a price control during month t. Sample restricted to each stage period. The regression follows equation (2), which includes category and time fixed effects. All coefficients are statistically significant at the 1% level (except the coefficient from stage 2, which is not statistically significant).

We also detect higher-priced varieties by looking at price dispersion before and after controls are introduced. This can be seen in figure 6b, which plots the price dispersion within subcategories for all goods and the subset of “continuing” goods.33 Panel b shows that price dispersion increases by around 14% during the first weeks after control. Furthermore, dispersion is primarily driven by new goods following price ceilings and does not revert to its initial levels. In both cases, price dispersion rises a few days before the price control is introduced, but in the sample that includes new varieties (all goods), the dispersion continues to rise after the control is in place.

We formalize the visual evidence of price dispersion using a similar regression to equation (2). In this case, the dependent variable is the category-URL price dispersion, defined as the coefficient of variation. The coefficient in column 4 in table 4 indicates that relative to the average price dispersion across all narrow categories, targeted price controls lead to 17.7% higher price dispersion.

VII. Differences across Stages

In table 5, we replicate the main empirical analyses for each of the price control stages to document their differences.

During the first stage, products under price controls experienced a temporary 0.78% decline in monthly inflation, a 27% probability of a stockout, and a 32% increase in price dispersion within subcategories, in line with the average across all stages.

Stage 2, the only nontargeted phase, was more successful in bringing aggregate inflation down temporarily (as shown in figure 2) but led to an increase of 2.37% monthly inflation for products that experienced targeted price controls at some other time. This suggests that the retailers utilized the freeze as an opportunity to anticipate (and possibly obfuscate) price increases in goods that they expected to be controlled in the future.

Stage 3 was the targeted phase with the strongest enforcement, which significantly reduced the inflation rate of targeted goods while simultaneously reducing the probability of a stockout. This was, not surprising, also the period when a large fraction of goods were discontinued, as seen in figure 4.

Finally, stage 4 appears to be the least successful phase of targeted controls, suggesting that the ability to enforce price controls had fallen dramatically by this time. This period also coincides with a new president in office, changes to the basket of controlled products, and potential right-censoring for controlled products introduced toward the end of the data collection.

VIII. Conclusion

During the past ten years, Argentina has experienced various forms of targeted price controls in which the government set price ceilings for specific supermarket goods. We use Web-scraping technologies to collect online prices from one of the largest retailers in the country, and we construct a detailed micro panel data set with more than 50,000 goods, which we use to evaluate the impact of price controls.

We show that although targeted price controls focused on goods with high CPI weight, they had minor and temporary effects on aggregate inflation. While these controls were binding in both price and product availability, firms introduced new varieties at higher prices to compensate for reduced profit margins.

Our results suggest that new technologies, such as the Internet and mobile phones, allow governments to better enforce targeted price control programs, but this does not make them an effective policy to reduce aggregate inflation. The effects are small and short-lived, and they do not spill over to noncontrolled goods. Furthermore, firms adjust to targeted price controls by using strategies that may obfuscate consumer options and increase price dispersion.

Future research should explore the welfare losses associated with additional search frictions, administrative enforcement costs, and price-obfuscation strategies in the retail industry.

Notes

1

Venezuela is a particularly troubling recent case of across-the-board price controls. The armed forces are in charge of strict price controls and a food supply program. In May 2017, President Maduro announced that he was considering a complete freeze of prices that would be enforced by “people in the streets” (El Nacional, 2015).

2

Targeted controls are also referred to as “selective controls” (Rockoff, 2004).

3

For example, developed countries such as Australia, Canada, and Switzerland currently have targeted price control programs for pharmaceutical drugs. See http://www.pbs.gov.au/pbs/home for Australia and http://www.pmprb-cepmb.gc.ca/home for Canada.

4

Rockoff (2004) provides an excellent overview of the history of price controls in the United States, ranging from the Revolutionary War to the Vietnam War. Galbraith (1952), Friedman (1966), and Solow (1966) provide early formal treatments on the subject. Helpman (1988) shows that the macroeconomic effects of price controls depend on market structure. Other papers include U.S. Council of Economic Advisers (1973), Gordon (1973, 1975, 1977), Schultze (1975), Darby (1976a), McGuire (1976), Oi (1976), Cox (1980), and Blinder and Newton (1981).

5

See FIEL (1990) for a review of price controls prior to 1990.

6

In addition to price controls, Argentina's main strategy to deal with inflation was to manipulate the official inflation statistics. See Cavallo (2013) for more details.

7

The new government elected in 2015 continued the Precios Cuidados program. More details can be found at the official government website: https://www.argentina.gob.ar/precios-cuidados. For evidence on their popularity, see Telam (2014b); Kollman (2014); Shaalo (2015); and Clarin (2016).

8

Section III describes how we identify controlled products. The appendix shows additional statistics for each of the control stages.

9

See Clarin (2013); Parks (2013). These efforts can be interpreted as a new form of bottom-up monitoring technologies. See also Ater and Rigbi (2018) for the effects of price transparency online.

10

As shown in section V, the ups and downs of price controls intensity are not related to data-scraping problems. Although price controls were supposed to be widespread during the second stage, retailers kept identifying a specific set of goods as being under a “government agreement.” Our matching algorithm, described in section III, identified these goods as “controlled products,” and therefore the intensity index is relatively stable during these weeks.

11

Several high-ranking government officials left around that time, including the secretary of commerce, the minister of economy, and the central bank president. See Sainz (2013b).

12

See Manzoni (2013), Sainz (2013a), and Clarin (2015).

13

The government monitored the retailers for stockouts, wrong labeling, incorrect prices, and incorrect product weight. The agreements also stipulated that retailers should not limit purchases of controlled goods per household. Companies and supermarket chains could be subject to temporary store closures and large monetary fines. For evidence on retailers receiving fines, see Buenos Aires Herald (2014); Telam (2014a); Lafuente (2017).

14

See La Nacion (2014).

15

The Israeli government has also faced challenges to enforce targeted price controls. See Globes (2017).

16

The product list can be found at https://www.argentina.gob.ar/precios-cuidados. Opposition leader President Mauricio Macri, who took office after former president Cristina Fernandez de Kirchner in December 2015, broadened the program's scope. See Telam (2016).

17

The same retailer is also used in Cavallo (2013, 2018). The retailer practices a uniform pricing strategy: online prices and availability are estimated to be identical across locations. Moreover, prices are estimated to be similar across the online and offline channels (Cavallo, 2018). In fact, the prices of controlled goods instructed by the government must be identical in both channels. We simulated online purchases and did not find that availability changed across locations. Although we lack data from a different retailer, we did not find news articles that reported differentiated enforcement across supermarket chains.

18

The appendix shows a screenshot of the government website.

19

This implies that controlled and uncontrolled products do not switch sample groups. A similar strategy is used in Cavallo (2013) and Rockoff (2004). Controlled goods in stage 2 are identified only through the HTML flag, although the government tried to impose a generalized freeze.

20

In the appendix, we expand on some of the key determinants of price controls and find that for a unit increase in the CPI weight (i.e., 1 percentage point) the odds of a control increase by 24%. Price controls are less likely in less concentrated or homogeneous markets, as approximated using the number of brands, products, and varieties within the category URL (i.e., narrow subcategories). These findings are consistent with the predictions in Cox (1980), who describes the policymakers' problem as maximizing the impact on the price index while reducing enforcement or deadweight costs.

21

See the appendix for additional evidence on the duration of price controls.

22

A body of literature argues that price controls can indirectly benefit the inflationary process through inflation expectations. See Galbraith (1952), Friedman (1974), Blinder and Newton (1981), and Rockoff (2004). Interestingly, inflation expectations remained relatively flat even around key price controls announcements. The appendix shows annual inflation rates, inflation expectations, and monetary policy. See the appendix also for visual evidence of higher volatility of controlled-goods' annual inflation rate, as well as a measure of “excess” inflation relative to noncontrolled sectors.

23

We focus on the first event to avoid picking up behaviors that are connected to previous controls, but the results are similar if we make no distinctions for repeated controls.

24

See also Blinder (1979) on related overpricing strategies during price controls in the Unites States.

25

Cavallo (2018) shows that in Argentina, close to 100% of the goods found offline are also available online and have similar prices.

26

A potential explanation is that price ceilings were being set above the intersection of demand and marginal cost curves in noncompetitive industries (Darby, 1976b; Helpman, 1988), so profit margins could be smaller but still positive.

27

These discontinuities explain why availability did not recover. It is also possible that as a new stringent, targeted program developed, firms preferred not to reintroduce controlled products for fear these would be the target of price controls again. In the appendix, we plot product introductions and discontinuities over time and show that these were more pervasive when the government increased the intensity of price controls.

28

Results remain robust to alternative functions, such as the nonparametric Nelson–Aalen cumulative hazard function (Kaplan & Meier, 1958).

29

Some firms actually benefited from the price agreements by gaining market share. In principle, these price agreements provided advertising and facilitated product distribution to major retailers throughout the country. For example, an Argentine firm producer of vinegar, mayonnaise, and other dressings reported that 27% of its 2014 sales could be attributed to the Protected Prices program and that these products exhibited a 28% increase in gross sales. Participating in the price agreements allowed the firm to access new retailers and supermarket chains in segments that were previously restricted to major brands. See Sainz (2015) and Telam (2015).

30

See Bourne (1919), Darby (1976a), Jonung (1990), and Rockoff (2004).

31

See also visual evidence of increased activity of product introductions and discontinuities in the appendix.

32

Parsing out the product description string into grams and liters per item, we find no evidence that firms systematically reduced package sizes to cope with price controls.

33

We compute price dispersion as the coefficient of variation, that is, standard deviation of prices over average prices, per week and URL. We then averaged these URL-level time series for each week, six months before and after the first control.

REFERENCES

Aparicio
,
D.
, and
M. I.
Bertolotto
, “
Forecasting Inflation with Online Prices,
International Journal of Forecasting
36
(
2020
),
232
247
.
Aparicio
,
D.
, and
R.
Rigobon
, “
Quantum Prices,
NBER working paper w26646
(
2020
).
Ater
,
I.
, and
O.
Rigbi
, “
The Effects of Mandatory Disclosure of Supermarket Prices
,”
CESifo working paper
6942
(
2018
).
Besanko
,
D.
,
S.
Donnenfeld
, and
L. J.
White
, “
Monopoly and Quality Distortion: Effects and Remedies,
Quarterly Journal of Economics
102
(
1987
),
743
768
.
Besanko
,
D.
,
S.
Donnenfeld
, and
L. J.
White
The Multiproduct Firm, Quality Choice, and Regulation,
Journal of Industrial Economics
36
(
1988
),
411
429
.
Blinder
,
A. S.
,
Economic Policy and the Great Stagflation
(
New York
:
Academic Press
,
1979
).
Blinder
,
A. S.
, and
W. J.
Newton
, “
The 1971–1974 Controls Program and the Price Level: An Econometric Post-Mortem,
Journal of Monetary Economics
8
(
1981
),
1
23
.
Bourne
,
H. E.
, “
Food Control and Price-Fixing in Revolutionary France: II,
Journal of Political Economy
27
(
1919
),
188
209
.
Buenos Aires Herald
, “
Supermarkets Fined with $6M over Price Watch Program,
” July 23,
2014
.
Cavallo
,
A.
, “
Online and Official Price Indexes: Measuring Argentina's Inflation,
Journal of Monetary Economics
60
(
2013
),
152
165
.
Cavallo
,
A.
Scraped Data and Sticky Prices,
” this review
100
(
2018
),
105
119
.
Cavallo
,
A.
, and
R.
Rigobon
, “
The Billion Prices Project: Using Online Prices for Measurement and Research
,”
Journal of Economic Perspectives
30
:
2
(
2016
),
151
178
.
Clarin
, “
Mirar para cuidar, control de precios,
Clarin
, May 29,
2013
.
Clarin
El Gobierno vigila etiquetas y packaging,
Clarin
, October 16,
2015
.
Clarin
El 25% de los consumidores busca los Precios Cuidados,
Clarin
, January 28,
2016
.
Cox
,
C. C.
, “
The Enforcement of Public Price Controls,
Journal of Political Economy
88
(
1980
),
887
916
.
Darby
,
M. R.
, “
The US Economic Stabilization Program of 1971–1974,
” in
M.
Walker
, ed.,
The Illusion of Wage and Price Control
(
Vancouver
:
Fraser Institute
,
1976a
).
Darby
,
M. R.
Price and Wage Controls: The First Two Years,
Carnegie-Rochester Conference Series on Public Policy
,
2
(
1976b
),
235
263
.
El Nacional
, “
Venezuelan President Puts Armed Forces in Charge of New Food Supply System,
El Nacional
, May 1,
2015
.
FIEL
,
Control de Precios e Inflacion: La Experiencia Argentina Reciente
(
Buenos Aires
:
Ediciones Manantial
,
1990
).
Friedman
,
M.
, “What Price Guideposts?” (pp.
17
39
), in
G. P.
Shultz
and
R. Z.
Aliber
, eds.,
Guidelines, Informal Controls, and the Market Place
(
Chicago
:
University of Chicago Press
,
1966
).
Friedman
,
M.
“Inflation, Taxation, Indexation” (pp.
71
81
), in
Institute of Economic
Affairs
, ed.,
Inflation: Causes, Consequences, Cures
(
London
:
Institute for Economic Affairs
,
1974
).
Galbraith
,
J. K.
,
A Theory of Price Control
(
Cambridge, MA
:
Harvard University Press
,
1952
).
Globes
, “
33% of Food Stores Violate Price Controls,
Globes
, August 13,
2017
.
Gordon
,
R. J.
, “
The Response of Wages and Prices to the First Two Years of Controls,
Brookings Papers on Economic Activity
1973
(
1973
),
765
779
.
Gordon
,
R. J.
The Impact of Aggregate Demand on Prices,
Brookings Papers on Economic Activity
1975
(
1975
),
613
670
.
Gordon
,
R. J.
Can the Inflation of the 1970s Be Explained?
Brookings Papers on Economic Activity
1977
(
1977
),
253
279
.
Gorodnichenko
,
Y.
,
V.
Sheremirov
, and
O.
Talavera
, “
Price Setting in Online Markets: Does IT Click?
Journal of the European Economic Association
16
(
2018
),
1764
1811
.
Helpman
,
E.
, “
Macroeconomic Effects of Price Controls: The Role of Market Structure,
Economic Journal
98
(
1988
),
340
354
.
Jonung
,
L.
,
The Political Economy of Price Controls: The Swedish Experience, 1970–1987
(
Farnham, UK
:
Gower
,
1990
).
Kaplan
,
E.
, and
P.
Meier
, “
Nonparametric Estimation from Incomplete Observations,
Journal of the American Statistical Association
53
(
1958
),
457
481
.
Kollman
,
P.
, “
Los Precios Cuidados, con amplio apoyo,
Pagina
, March 10,
2014
.
Kyle
,
M. K.
, “
Pharmaceutical Price Controls and Entry Strategies,
” this review
89
(
2007
),
88
99
.
La Nacion
, “
Cristina Kirchner llamo a una consumidora que se quejo en Facebook por Precios Cuidados,
La Nacion
, February 6,
2014
.
Lafuente
,
A.
, “
Multan a los supermercados por $18 millones por incumplimientos en la exhibicion de precios y otras faltas,
La Nacion
, June 1,
2017
.
Leffler
,
K. B.
, “
Ambiguous Changes in Product Quality,
American Economic Review
72
(
1982
),
956
967
.
Manzoni
,
C.
, “
La yerba mate intenta esquivar a Moreno con mas variedades,
La Nacion
, September 1,
2013
.
McGuire
,
T. W.
, “
On Estimating the Effects of Controls,
” in
Carnegie-Rochester Conference Series on Public Policy
2
(
1976
),
115
156
.
Nixon
,
R. M.
,
RN: The Memoirs of Richard Nixon
(
New York
:
Grosset & Dunlap
,
1978
).
Oi
,
W. Y.
, “
On Measuring the Impact of Wage-Price Controls: A Critical Appraisal
” (
2:7–64
), in
Carnegie-Rochester Conference Series on Public Policy
,
North-Holland
,
1976
).
Parks
,
K.
, “
Argentina Takes Another Shot at Price Controls,
Wall Street Journal
, December 20,
2013
.
Raymon
,
N.
, “
Price Ceilings in Competitive Markets with Variable Quality,
Journal of Public Economics
22
(
1983
),
257
264
.
Raszewski
,
E.
, “
Argentine Price Controls Spark Milk, Meat Shortages,
Bloomberg
, April 16,
2007
.
Raszewski
,
E.
Argentine Markets Freeze Prices for 60-Days Amid 26% Inflation,
Bloomberg
, February 4,
2013
.
Rockoff
,
H.
,
Drastic Measures: A History of Wage and Price Controls in the United States
(
Cambridge
:
Cambridge University Press
,
2004
).
Sainz
,
A.
, “
Burlan el congelamiento de precios con nuevas marcas y cambios de formula y tamano,
La Nacion
, June 3,
2013a
.
Sainz
,
A.
Dos dias antes del vencimiento, no se sabe si seguira la lista de 500 productos congelados,
La Nacion
, October 30,
2013b
.
Sainz
,
A.
Precios cuidados: un exito de marketing, un fracaso para frenar la inflacion,
La Nacion
, May 20,
2015
.
Schultze
,
C. L.
, “
Falling Profits, Rising Profit Margins, and the Full-Employment Profit Rate,
Brookings Papers on Economic Activity
1975
(
1975
),
449
471
.
Shaalo
,
M.
, “
Precios Cuidados representa apenas el 10% de las ventas en super-mercados,
El Cronista
, July 20,
2015
.
Solow
,
R. M.
, “The Case against the Case against the Guideposts” (pp.
41
54
), in
G. P.
Shultz
and
R. Z.
Aliber
, eds.,
Guidelines, Informal Controls, and the Market Place
(
Chicago
:
University of Chicago Press
,
1966
).
Telam
, “
Clausuran en La Plata dos supermercados por incumplir con el programa Precios Cuidados,
Telam
, November 3,
2014a
.
Telam
Precios Cuidados: se cuadriplico la facturacion de productos en los super,
Telam
, October 17,
2014b
.
Telam
Augusto Costa asocio la inversion de una empresa alimenticia con la participacion en el programa Precios Cuidados,
Telam
, March 20,
2015
.
Telam
Se extiende Precios Cuidados, con 530 productos y subas promedio de 2.49%,
Telam
, September 6,
2016
.
U.S. Council of Economic Advisers
(
1973
):
Economic Report of the President
(
Washington, DC
:
U.S. Government Printing Office
,
1973
).

External Supplements

Author notes

We thank the editor, Olivier Coibion, as well as four anonymous referees, for very helpful suggestions. We benefited from comments by Glenn Ellison, Manuel Bertolotto, Vivek Bhattacharya, Carolina Levinton, Ignacio Puente, Roberto Rigobon, Paulo Somaini, Jean Tirole, and workshop participants at MIT. The data and code needed to reproduce the results in this paper are publicly available at http://www.thebillionpricesproject.com.

A supplemental appendix is available online at https://doi.org/10.1162/rest_a_00880 or on the authors' web pages.

Supplementary data