Macro Roundup

Daniel Garcia-Macia, Chang-Tai Hsieh and Peter Klenow, "How Destructive Is Innovation," Econometrica, http://klenow.com/DestructiveInnovation_GHK.pdf

Bottomline, “..How much innovation takes the form of creative destruction versus new varieties versus firms improving their own products? How much occurs through entrants versus incumbents? We try to infer the sources of innovation from the employment dynamics of U.S. firms in the nonfarm private sector from 1983 to 2013. We conclude that creative destruction is vital for understanding job destruction and accounts for around one-fourth of growth. Own-product quality improvements by incumbents appear to be the biggest source of growth. Net variety growth contributes much less than quality improvements doOur findings are relevant for innovation policy. According to Atkeson and Burstein (forthcoming), the consumption-equivalent welfare gain from devoting about 1% more of GDP to research every year is between 0.17 and 0.73 percent in a model calibrated similarly to ours (but with endogenous research investments). They estimated the gains would be larger—0.26 to 2.01 percent—if creative destruction played no role in U.S. growth. Creative destruction is a force raising the private return relative to the social return to research, diminishing the gains from promoting research….”

“…. Table IV presents TFP growth due to creative destruction (row 1), new varieties (row 2), and own innovation (row 3). TFP growth due to each source of innovation is the product of arrival rate of innovation and the quality improvement conditional on innovation. We use equation (2) for this calculation.18 The first column shows that TFP growth due to own innovation was about 1% per year in 1983-1993. Growth due to creative destruction was about half that, at 0.44% per year. And new varieties generate growth of 0.23% per year. The rows in Table V show the contribution of each source of innovation to aggregate TFP growth. About 27% of the 1.66% growth rate in the 1983-1993 period comes from creative destruction. Own-variety improvements by incumbents account for 60%. New varieties à la Romer (1990) are the remainder at around 14%.The columns in Table V also decompose aggregate TFP growth into the percentage contribution of entrants versus incumbents using equation (3). In the ten years between 1983 and 1993, incumbents account for 68% of aggregate TFP growth, with entrants contributing the remaining 32%. Aghion, Akcigit, and Howitt (2014) provided complementary evidence for the importance of incumbents based on their share of R&D spending and patents…”

TFP driver estimates, “Table III presents the eight parameter values inferred from the data using the procedure described above. Based on the data moments from 1983 to 1993, we infer a 70% arrival rate of own-variety quality improvements per 5-year period. Conditional on no own innovation, quality improvements through creative destruction occur 30% of the time by other incumbents. Conditional on no own innovation and creative destruction by another incumbent, quality improvement through creative destruction by entrants occur with probability 1. The unconditional probability that a given product improves due to creative destruction by an incumbent is thus 8.9%, and the unconditional probability of creative destruction by an entrant is 21.0%.16 The unconditional probability that a product is improved upon in a 5-year period is thus 100%, of which 69% is from own innovation and 31% is from creative destruction (the latter from entrants or incumbents). The employment-weighted average step size for quality improvements on existing varieties is given by sq = (θ/(θ − (σ − 1)))1/(σ−1). Given that θ = 154 and σ = 4, the average improvement in quality (conditional on innovation) is 7.5%.17 New varieties are only created by entrants, arrive with 12.3% probability per existing variety, and have an average quality that is 31% of the average quality of existing varieties. Overhead costs imply that the average quality of exiting products ψ is 2% of the average quality of existing varieties, and that the probability a variety exits due to overhead cost δo is essentially zero. The net number of varieties thus grows by 12.3% every five years, which matches the growth of total employment and number of firms from 1983 to 1993.

Core of paper, “…Entrants and incumbents can create new products and displace the products of competitors. Incumbents can also improve their existing products. How much of aggregate productivity growth occurs through each of these channels? Using data from the U.S. Longitudinal Business Database on all nonfarm private businesses from 1983 to 2013, we arrive at three main conclusions:First, most growth appears to come from incumbents. We infer this from the modest employment share of entering firms (defined as those less than 5 years old). Second, most growth seems to occur through improvements of existing varieties rather than creation of brand new varieties. Third, own-product improvements by incumbents appear to be more important than creative destruction. We infer this because the distribution of job creation and destruction has thinner tails than implied by a model with a dominant role for creative destruction…”

Colin Hottman, Stephen Redding, David Weinstein, "Quantifying the Sources of Firm Heterogeneity,"The Quarterly Journal Of Economics, March 2016, https://academic.oup.com/qje/article-abstract/131/3/1291/2461153

Their takeaway, ““…New and improved products (scope and turnover) account for 64% of firm growth in our three highest turnover sectors, but only 10% of firm growth in the three lowest turnover sectors. Although product upgrading alone accounts for 15% of firm growth in the three highest turnover sectors, it only accounts for 0.2% of firm growth in our three lowest turnover sectors. Taken together, these decomposition results are quite striking in their consistency about the sources of firm heterogeneity.Regardless of whether we examine firms in the cross section or in the time series, improvements in firm appeal followed by product scope are the most important drivers of firm sales for small and large firms alike. The role played by average marginal cost differences tends to be small and depend on the sample and decomposition method used. But we consistently find a positive contribution towards differences in firm size from the cost dispersion term. Finally, we see that although markup differences are unimportant determinants of firm sales for most firms, they do appear to be important for the very largest firms, something that suggests aggregate implications from the exploitation of market power by these firms……Our findings that firms supply multiple imperfectly substitutable varieties have important implications for the measurement of firm productivity, highlighting the role of assumptions about demand in the measurement of productivity for multiproduct firms. Conventional price indexes based on a weighted average of firm prices do not take into account that the theoretical price index for the firm depends on the number of products it supplies whenever consumers care about variety. Indeed, our counterfactual exercise indicates that the multiple varieties supplied by multiproduct firms reduce the aggregate consumer price index by around one third. Moreover, ignoring this effect introduces a systematic bias into the measurement of firm productivity, because larger firms supply more products than smaller firms. We find this bias to be quantitatively large. An increase in firm sales is associated with around a one third larger increase in true real output (using the exact price index) than in measured real output (using the conventional price index)…”

“.. The extent of multiproduct firms can be seen more clearly in Table IV, which shows the results of splitting the data by the number of UPCs supplied by a firm. Although single-product firms constitute about one third of all firms on average, these firms account for less than 1% of all output. In other words virtually all output is supplied by multiproduct firms. Moreover, the fact that over 90% of output is sold by firms selling 11 or more varieties and nearly two thirds of all output is supplied by firms selling more than 50 varieties suggests that single-product firms are more the exception than the rule. Large firms not only sell more products but also a lot more of each product. The penultimate column of Table IV documents that while the typical barcode sold by a single-product firm brings in $63,871 in revenue, the typical barcode sold by a firm selling over 100 barcodes brings almost twice as much ($122,045). In other words, large firms not only supply more products, they sell more of each product. If firms differed only in the fixed cost of adding new varieties, one would not expect to see large firms sell more of each variety. The fact that they do strongly suggests that large firms must also differ in the marginal cost or demand for their output. Finally, although the largest firms have nontrivial shares of particular product groups, these firms are small compared to the U.S. economy. Ninety-nine percent of firms in our sample have aggregate market shares (across all product groups) of less than 0.1% of total barcode sales. Even the largest firm only sells 3% of total barcode sales in our sample. Given that sales of packaged goods is only a fraction of total U.S. sales in all sectors, it is reasonable to conclude that no individual firm has the capacity to affect aggregate U.S. prices, expenditure, or welfare….”

“…We see in Table II that almost 90% of sales in a product group was produced by firms with sales in the top decile of sales.Table III provides a more detailed description of this firm heterogeneity by focusing on the 10 largest firms in each product group (where we weight the averages by the sales of the product group). Table III reveals an almost fractal nature of firm sales. Around two thirds of all of the sales of firms in each product group is produced by the 10 largest firms (which on average only account for 2% of firms in each product group). While on average half of all output in a product group is produced by just five firms, 98% of firms have market shares of less than 2%. Thus, the typical sector is characterized by a few large firms and a competitive fringe composed of firms with trivial market shares. A second striking feature of the data is that even the largest firms are not close to being monopolists. The largest firm in a product group on average only has a market share of 22%. Finally, the data reveal that firms in the top decile of sales are all multiproduct firms, supplying on average 68 different goods with the largest firms supplying hundreds of goods…”

The Evidence, “…One of the most striking facts displayed in this table is the degree of firm heterogeneity. This is manifest in the skewness of the size and barcode distributions. The largest firm in an industry typically sells 2,500 times more than the median firm. We see similar patterns in terms of product scope and sales per product. The firm with the most products typically has 97 times more products than the firm with the median number of barcodes, and the barcode with the most sales on average generates almost 900 times more revenue than the revenue of the median barcode…”

On large firms productivity relative to small firms, “…These results imply that about a quarter of the variation of firm-level real output one would obtain by using a conventional price index is simply due to the fact the conventional price index assumes that firms produce homogeneous output—an assumption that easily can be rejected. Moreover, studies based on conventional measures of firm prices understate the real output of large firms by a third relative to small firms, with implications for estimates of returns to scale and productivity….”

Core results, “…Our results point to differences in firm appeal as being the principal reason some firms are large and others are not. Depending on the specification considered,we find that 50-75% of the variance in firm size can be attributed to differences in appeal, about 20-25% to differences in product scope, and less than 20% to average marginal cost differences. If we use a broad measure of total firm appeal, which encompasses both firm appeal as well as scope, we find that total firm appeal accounts for almost all of firm size differences. We estimate substantially higher elasticities of substitution between varieties within firms than between firms (median elasticities of 6.9 and 3.9, respectively), implying that a firm’s introduction of new product varieties cannibalizes the sales of existing varieties. We estimate that the cannibalization rate for the typical firm is 0.50, roughly halfway between the extreme of no cannibalization (equal elasticities of substitution within and between monopolistically competitive firms), and the extreme of complete cannibalization (varieties perfectly substitutable within firms). We find that the typical sector comprises a few large firms with substantial market shares and a competitive fringe of firms with trivial market shares. Therefore most firms charge markups close to the monopolistic competition benchmark of constant markups, because they have trivial market shares and hence are unable to exploit their market power. However, the largest firm accounts on average for 22% of sales in a sector, and the median largest firm charges a markup between 24% and 100% higher than the average firm within the same sector. Using the estimated model to undertake counterfactuals, we find that these departures from the monopolistically competitive benchmark raise aggregate consumer price indexes by between 4% and 13%...”

Key quote, “…Our results point to demand differences (which could arise from quality or taste variation) as being the principal reason some firms are successful in the marketplace and others are not. Depending on the specification considered, we find that 50- 70% of the variance in firm size can be attributed to differences in firm appeal, about 20-25% to differences in product scope, and less than 25% to cost. When we turn to examine time-series evidence, the results become even more stark. Virtually all firm growth can be attributed to firm appeal, with most of the remainder due to product scope. These results suggest that most of what economists call differences in revenue productivity reflects differences in appeal (e.g., quality or taste) rather than cost….“…We show that if demand has a nested CES structure, conventional measures of real output will have a downward bias that rises with firm size with an elasticity of around one third. In other words, real output variation is substantially greater than nominal output variation. This bias also implies that true productivity differences are much larger than conventionally measured ones…”

Core of paper, “…Returning to our baseline estimates, Table 3 compares our missing growth to official TFP growth. The entries are annual percentage points. As mentioned, we find 0.54 percentage points of missing growth per year from 1983-2013.BLS measured TFP growth over the same interval was 1.87 percentage points per year.24 If we add our missing growth to the BLS TFP series we arrive at “true” growth of 2.41 percent per year. Thus, our baseline estimate is that over one-fifth of true growth is missed. Table 3 also breaks the 30-year sample into three subperiods: 1983-1995 (an initial period of average official growth), 1996-2005 (a middle period of rapid official growth), and 2006-2013 (a final period of low official TFP growth). Did missing growth contribute to the speed-up or slow-down? Our estimates say yes, but modestly at most. Missing growth slowed down by 4 basis points when official growth accelerated by 88 basis points in the middle period. And missing growth sped up by 17 basis points when official growth dropped 170 basis points in the final period. While missing growth did not accelerate significantly in the aggregate, it did so in the Information sector. In this sector, missing growth rose from 0.22 percent per year over 1983-2013 to 0.90 over 1996-2005 and to 1.90 over 2006-2013. Due to its small employment share, however, the Information sector contributed a small amount to overall missing growth: 0.00 over 1983-2013, 0.02 over 1996-2005, and 0.04 over 2006-2013….”

Impact on IT, Ben notes, “..it accelerates a lot so the average isn’t high but it gets bigger through time…”“…. While missing growth did not accelerate significantly in the aggregate, it did so in the Information sector. In this sector, missing growth rose from 0.22 percent per year over 1983-2013 to 0.90 over 1996-2005 and to 1.90 over 2006-2013. Due to its small employment share, however, the Information sector contributed a small amount to overall missing growth: 0.00 over 1983-2013, 0.02 over 1996-2005, and 0.04 over 2006-2013…”

Bottomline, "...Our findings can be summarized as follows. First, missing growth from imputation is substantial: roughly one-half a percentage point per year, or around one third of measured productivity growth. Second, missing growth is concentrated in hotel, restaurants, and retail trade rather than manufacturing. Third, missing growth accelerated only modestly after 2005, an order of magnitude smaller than needed to explain the slowdown in measured growth...."
Core results, "...Based on the model and US Census data for all nonfarm businesses, we estimated the magnitude of missing growth from creative destruction over the period 1983-2013. Our approach uses the market share of surviving, entering, and exiting plants. We found (i) missing growth from imputation was substantial at around one-half of a percentage point per year, or over one-fifth of measured productivity growth; (ii) missing growth was concentrated in hotels, restaurants, and retail trade rather than manufacturing; and (iii) it has accelerated modestly since 2005, so that it accounts for only about one-tenth of the sharp growth slow-down since the..."

The Evidence, “…Our baseline results calculate missing growth from 1983-2013 using LBD data. The LBD contains data on employment and payroll going back to 1976, but the payroll data feature a number of implausible outliers before 1989. We therefore use employment data for our baseline estimates. Using the employment data, we identify entrants beginning in 1977; 1983 is the earliest year we can calculate missing growth (market share growth of survivors between 1982 and 1983) because we use plants that have been in the data for at least five years (1977 to 1982 at the beginning). We examine missing growth starting in 1983.18 In the fourth column of Table 1, we report our missing growth estimates for each two-digit sector using Hall’s (2018) implied σ. We order the sectors by their contribution to missing growth over the period 1983-2013, which totals 54 basis points per year.19 The biggest contributors, by far, are NAICS 72 (hotels and restaurants) and 44-45 (retail trade); they contribute 33 of the 54 basis points, or over 62 percent of the total.Their contribution may reflect the geographic spread of outlets by big national chains. No other sector contributes more than 5 basis points.20 The dominant role played by hotels, restaurants, and retail trade hearkens back to studies finding that large entering establishments carried productivity growth in these sectors. See Foster, Haltiwanger, and Krizan (2006) and Jarmin, Klimek, and Miranda (2009) for evidence on retail trade, and Hanner et al. (2011) for documentation of the big-box revolution among grocery stores, mass merchandisers, and restaurants. We stress that the growth generated by this revolution may not have been fully captured in official statistics. Perhaps the most surprising result in Table 1 is the small contribution of manufacturing. Within this sector we estimate missing growth of about 5 basis points per year, so that it contributes less than 1 basis point per year to overall missing growth. By focusing on domestic production, we are overlooking the gains in import variety emphasized in the trade literature…”

Max Eden critique of study of Boston’s public pre-K program, “….It would still be far from clear that this study proves that universal pre-k would be a wise public investment. That’s because policymakers must necessarily deal in tradeoffs and counterfactuals, and this study leaves unanswered the question about what would have happened to student outcomes if instead of launching a pre-K program, Boston had instead simply given families $20,000 per child.”…”

Max Eden, "Boston pre-K miracle or mirage: What can the latest pre-K study tell policymakers?," American Enterprise Institute, May 25, 2021, https://www.aei.org/op-eds/boston-pre-k-miracle-or-mirage-what-can-the-latest-pre-k-study-tell-policymakers/

Boston pre-K miracle or mirage: What can the latest pre-K study tell policymakers?

Universal pre-K advocates have historically used studies the way a drunk man uses lampposts - for support rather than illumination, as the Brookings Institution’s Russ Whitehurst has pointed out. Their argument could be justly paraphrased: We have strong evidence that small, intensively resourced programs serving deeply disadvantaged students yielded strong benefits; therefore, we know that large-scale, less resourced, differently designed universal programs will also yield strong benefits.

But just as Congress is debating President Biden’s $1.8 trillion American Families Plan, with $425 billion going towards universal public pre-K as well as universally affordable child care, researchers have released a discussion draft of a paper demonstrating, as an author told the New York Times, “that modern large-scale public preschool programs can improve educational attainment.”

The study examines long-term outcomes for students chosen randomly by lottery to attend Boston’s public pre-K program (for four-year-olds) from 1997 to 2003. The top-line results are strikingly positive: participants were 6 percentage points more likely to graduate high school, and 5.4 percentage points more likely to enroll in college. They also received fewer suspensions in high school and were less likely to be incarcerated as a juvenile.

But underneath that topline, the findings get somewhat peculiar. In theory, public pre-k should provide a benefit either by allowing kids to enroll in a higher quality program than their families could afford, or by providing them with a more developmentally congenial environment than they would experience at home. Pre-K should, then, provide greater benefits for disadvantaged students from single-parent households. Much of the literature on childcare and pre-K conforms to this expectation.

But not this study. To put it somewhat crudely, but also probably more accurately than the authors put it themselves, this study finds that the benefits of pre-K are largely the preserve of affluent white boys.

In their subgroup analysis by gender (see Table 8 in the study), the authors found no benefits for girls significant at a 5% or higher level. They found benefits for boys significant at a 5% or higher level for: taking the SAT test, high-school graduation, discipline, four-year-college enrollment, and college graduation. By race, white students saw significant improved outcomes on college enrollment, graduation, and standardized math test scores; black students saw improved outcomes only on discipline and SAT-taking; Hispanic students saw an increase in on-time college enrollment, but (oddly) a substantial deterioration in standardized math test scores. By poverty, higher-income students saw gains in college enrollment, college graduation, and standardized math test scores. Low-income students saw benefits only for discipline and SAT-taking.

But for the moment, let’s stipulate that this discussion draft is perfectly internally valid. In terms of external validity, it is surely an order of magnitude superior to the “go-to” studies typically touted by early education advocates (the High/Scope Perry Preschool Project and the Carolina Abecedarian Project). But it is still an extremely thin and tenuous reed on which to make the case for rapidly scaling universal pre-K for 3- and 4-year-olds with a massive infusion of federal cash.

The researchers claim that Boston’s pre-K program costs about $13,000 per child for a full-day, full-year program in today’s dollars. But that is a sloppy and incorrect cost calculation, based not on actual administrative expenses but rather on a third-party projection made two years prior to the launch, which assumed that middle-class parents would pay substantial fees to access public pre-K. Given that Boston pre-K teachers “are paid on the same scale as K-12 teachers that pre-k teachers are required to get a masters’ degree within five years,” and that pre-K class size is similar to class size in the public school system, a more reasonable back-envelope calculation puts the cost of the program in line with overall per-pupil spending of about $20,000.

Furthermore, although Boston’s pre-K program was technically “universal” insofar as it was open to students regardless of income, it was still relatively small during the time period studied. At launch in 1997, Boston’s program served more than 2,500 students. In 1998, the school district cut the program back to under 1,000, where it remained for the duration of the period that was studied. Cutting back dramatically on the program may have enabled the school to select only the best teachers to remain in it, further undercutting the external validity of the study’s findings. While the program has grown since, the authors note that today—almost 25 years after—the pre-K program only has capacity to serve roughly half of Boston’s pre-K students.

When it comes to assessing the likely effect of a rapid and massive expansion of pre-K intended to yield a universal entitlement, the studies with the greatest external validity remain those conducted on the Quebec Family Plan and the Tennessee Voluntary Pre-K (VPK) Program. Both provide grounds for substantial alarm.

Quebec’s rapid and massive expansion of affordable child care/pre-K, which yielded a 14-percentage point increase in the share of 0-4 year-olds in center-based care, did substantial and lasting harm to children. Researchers found that time in child care was linked to increased hyperactivity, anxiety, and aggression, as well as deterioration in motor and social development. Research also found strong evidence of “worse parenting” after the new policy was put into place, as well as a significant deterioration in the reported quality of the relationship between the child’s parents. Based on all this, the researchers concluded that “the consistency of the results suggest that more access to child care is bad for these children (and, at least along some dimensions, for these parents).” Further studies provided the small silver lining that this damage was confined to students from two-parent households, and that child care still did some good for children from single-parent households. But the damage done was lasting. Researchers concluded that “the negative impact of the Quebec program on the noncognitive outcomes of young children appears to persist and grow as they reach school ages,” including a rise in anxiety and aggression as well as worse health and lower life satisfaction.

Tennessee’s statewide voluntary pre-K program serves about 18,000 low-income children at a high level of quality, according to the National Institute of Early Education Research’s metrics, but at a cost of less than $5,000 per student. The results were bad: by third grade, VPK students performed substantially worse on science and math tests, were more likely to be disciplined for behavioral infractions, and more likely to be diagnosed with disabilities.

The Boston Pre-K study provides evidence adequate to reasonably advise city mayors or school boards of well-resourced districts to experiment with the provision of small-scale pre-K programs. But as Congress considers dramatically expanding investment in early education, the Quebec Family Study and the Tennessee VPK study hold vastly more relevance to the American Families Plan, which would (as in Quebec) dramatically scale up early education provision and likely sustain it at a cost point closer to Tennessee’s $5k per pupil than Boston’s $20k. Hence we still, on balance, have more reason for alarm than optimism about what Biden’s American Families Plan will mean for children.

But for the sake of argument, let’s wave away the very real concerns about implementation. And let’s assume that the American Families Plan would yield two more grades in public schools funded at around $20k per year. And let’s assume that from this investment we could expect benefits similar to what researchers found in Boston. And let’s assume that these benefits would be universal, not reserved for affluent white boys.

It would still be far from clear that this study proves that universal pre-k would be a wise public investment. That’s because policymakers must necessarily deal in tradeoffs and counterfactuals, and this study leaves unanswered the question about what would have happened to student outcomes if instead of launching a pre-K program, Boston had instead simply given families $20,000 per child.