We now turn to our central question: what are the drivers of upgrading? I categorize drivers into three groups, which can be understood with reference to the general framework above: (1) output-side drivers: factors that affect product demand curves (the D(·) functions); (2) input-side drivers: factors that affect input-supply curves (the S(·) functions); and (3) drivers of capabilities: factors that affect the the “know-how” of firms (the Λijkt, Jit, and Kijkt). This categorization is necessarily somewhat loose - some drivers fit in more than one category, and some not quite in any - but the grouping is helpful to organize the review.
3.3 Drivers of Firm Capabilities
This section reviews research on factors that operate through effects on firm capabilities and knowledge. A first issue that arises is the motivation of entrepreneurs, in particular whether or not they can be presumed to maximize profits. We then turn to various factors that influence firms’ know-how.
3.3.1 Objectives of Entrepreneurs
The framework in Section 2.1 assumes that firm seeks to maximize the discounted present value of profits, expressed in equation (2). Is this a plausible assumption? One reason it may not be is that entrepreneurs consciously hold other objectives. Entrepreneurs may value a quiet life (Bertrand and Mullainathan, 2003) or derive private benefits from control or empire-building (Williamson, 1964). Although these motivations are often attributed to non-owner managers, they might also characterize owners themselves. Another possible reason is that entrepreneurs would like to maximize profits but have behavioral biases that lead them to make mistakes. While these possibilities are widely acknowledged, there is relatively little empirical research directly on the question of whether individual owners of medium-sized or large firms hold non-profit-maximizing objectives or systematically make mistakes.[1] There is evidence suggesting that mistakes are made by small shopkeepers, in the form of lost sales due to holding insufficient change (Beaman et al., 2014), and by agricultural producers, in the sense of failing to notice relevant information about production (Hanna et al., 2014) or failing (because of time-inconsistent preferences) to invest in fertilizer (Duflo et al., 2011). But more empirical investigation of the objectives consciously held by firm-owners and of their behavioral biases is sorely needed.
Two words of caution are in order. First, the question of whether an individual entrepreneur maximizes utility is distinct from the question of whether a firm profit-maximizes. As we will see below, a firm may fail to take advantage of an apparent profit-making opportunity, even if all individuals within the firm are behaving rationally, in pursuit of standard objectives. Second, it appears to have become more common in recent years to attribute poor firm performance in developing countries to failures of entrepreneurs to profit-maximize. But as noted above, entrepreneurs in developing countries often face very different conditions in product and input markets, and hold different amounts of know-how, from rich-country entrepreneurs. We need to examine very closely the constraints they face before we can conclude that they have failed to optimize. In an agricultural context, Schultz (1964), Stiglitz (1989) and others have argued for a “poor but rational” view: if we observe behavior that seems to be non-optimal, we should ask ourselves what problem is being solved, and what constraints producers face, before concluding that they are not optimizing. A similar point applies to entrepreneurs in larger manufacturing firms. This is not to say that all developing-country entrepreneurs are perfect exemplars of Homo Economicus, but rather that we should be cautious before concluding that they are not.
3.3.2 Entrepreneurial Ability
Turning to drivers of capabilities, a first one to consider is entrepreneurial ability, which we can think of as a fixed characteristic of an individual entrepreneur - in the framework of Section 2.1, a time-invariant component of capability that is common across products and techniques. Recent research has taken several approaches to evaluating the importance of entrepreneurial ability. One approach is to examine cross-sectional correlations between detailed manager characteristics and firm performance. For instance, there is evidence from a range of countries, including Brazil and India, that firm performance is positively correlated with the amount of time CEOs spend in high-level meetings, rather than production activities (Bandiera et al., forthcoming). Focusing on six factories of an Indian garment firm, Adhvaryu et al. (2019a) find that factor-analytic summary measures they characterize as managerial attentiveness and autonomy correlate positively with levels of productivity and the rate of productivity improvement on production lines.[2] A natural question that remains open is whether the correlations reflect causal effects of manager characteristics or some form of sorting of managers to firms or production lines.
Another way to assess the role of such fixed manager characteristics is to examine changes in firm decisions and performance in response to changes in top managers. This is the strategy of Bertrand and Schoar (2003), who find in US data that manager fixed effects have significant explanatory power for various corporate decisions, even controlling for rich sets of firm observables.
A small literature examines the decisions and performance of family-owned firms where managerial positions are passed between family members (as opposed to being filled through competitive searches). There is robust evidence that inherited control is bad for performance (Pérez-González, 2006; Bennedsen et al., 2007; Bertrand et al., 2008). There is also evidence that family control is associated with lower scores on the World Management Survey index (Bloom and Van Reenen, 2007, 2010; Bandiera et al., 2017). Instrumenting family control with the gender mix of the previous CEOs’ children, Lemos and Scur (2019) have recently shown that this relationship is causal: family control leads to lower-scoring management practices.
Another type of evidence comes from changes of ownership. Using detailed data on ownership and physical inputs and outputs in the Japanese cotton spinning industry in the Meiji era, Braguinsky et al. (2015) find that acquisitions are associated with increases in TFPQ in the acquired firms. Interestingly, the acquiring firms typically do not have higher physical productivity than the acquired firm prior to purchase, but they are more profitable, in part, the authors suggest, because they are able to manage demand fluctuations to maintain higher levels of capital utilization. Using a propensity-score matching estimator in Spanish data, Guadalupe et al. (2012) find that acquisition by a multinational firm leads to upgrading on a number of directly observable dimensions, including indicators for process and product innovations, purchases of new machinery, and the introduction of new organizational practices. Studies in developing countries have largely found positive effects of foreign ownership on productivity (Arnold and Javorcik, 2009; Javorcik and Poelhekke, 2017; Stiebale and Vencappa, 2018), although there is still a debate about whether acquisition by multinationals has larger impacts than acquisition by domestic firms (Wang and Wang, 2015). In Indian data, Stiebale and Vencappa (2018) also find evidence of a positive effect of foreign acquisition on quality upgrading, indicated both by an increase in input prices and by a measure of product quality along the lines of Khandelwal et al. (2013).
Overall, the evidence seems strong that entrepreneurial ability matters for upgrading outcomes and that family control is associated with worse performance. This raises a question of why family control is so prevalent, a topic to which we return in the next subsection.
3.3.3 Agency Issues
Firms are collections of people with sometimes aligned but sometimes conflicting interests. Even if an entrepreneur is rational and of high ability, she may still have difficulties in getting employees to act in a desired way. These agency issues can be thought of as influencing a firm’s capabilities. The extent to which a firm is able to resolve them will clearly matter for its ability to upgrade. The agency literature is very large;[3] here we focus on empirical studies in developing countries on how agency issues influence upgrading outcomes at the firm level.
The Atkin et al. (2017b) study of Pakistani soccer-ball producers highlights the importance of such agency issues. Through a series of fortuitous events, the research team came up with a new technology - a design for cutting more pentagons from a rectangular sheet and a piece of equipment, an “offset” die, to implement that design. An advantage of the context is that all firms use the same, simple production process, at least for part of their production, and it is possible to calculate directly the benefits of adoption, which are positive on net for essentially all firms.[4] The researchers gave out the technology to 35 firms, expecting the treated firms to adopt quickly and planning to track the channels of spillovers. But 15 months later, only 5 treated firms and 1 control firm had adopted, despite the fact that the technology appeared to be working as expected. Conversations with firm owners and employees revealed the reason: the key employees, cutters, were paid piece rates based on the number of pentagons cut, with no incentive to reduce waste, and the offset die slowed them down initially. Although the reductions of waste were much larger than the increases in labor costs, under the existing contracts the cutters’ incomes would have declined with adoption and so they found various ways to discourage it. The researchers conducted a second experiment in which employees received a bonus of a month’s salary if they demonstrated the productivity benefits of the new die in the presence of their employer. The second experiment generated a statistically significant increase in adoption by firms, suggesting that a conflict of interest within the firm had been at least in part responsible for the initially slow adoption of the offset dies. A natural question is why firm owners did not adjust their payment schemes to reward the employees for adopting the new technology (or at least keep them whole). One possibility is that owners simply did not understand the problem; another, consistent with qualitative evidence, is that they understood, but that there are costs to changing employment contracts, even informal ones, and that owners calculated (perhaps with low priors about the value of the technology) that the expected benefits did not compensate for the re-contracting costs. The failure to adopt the new dies is arguably an example of what Garicano and Rayo (2016) call an “organization failure” - the firm as a whole failed to adopt a more-efficient technology - even though all individuals in the firm appear to have been acting rationally, given their knowledge. The case is also arguably an example where contracts that were optimal in a technologically static environment (here, piece rates before the new die) were not optimal in a technologically dynamic one (once the new die was introduced), and the stickiness of contracts generated a sort of organizational inertia.
A recent study of the adoption of credit scoring by Indian banks by Mishra et al. (2019) provides additional evidence for organizational inertia. The key finding is that older banks, both public and private, founded prior to the beginning of India’s liberalization in 1991, are less likely to adopt credit scoring for existing clients than the same banks are for new clients or than new banks (founded post1991) are for existing clients. The authors suggest that the older banks developed an organizational culture and way of dealing with existing clients under the less competitive pre-liberalization regime and that the culture has persisted, fading away only slowly.
A recent study by de Rochambeau (2017) identifies another sort of agency issue. The author randomly gave out GPS monitors to trucking firms in Liberia. She finds that they reduced unauthorized breaks and average travel times for the trucks on which they were installed, as expected. But she also finds that owners were less likely to install the monitors on trucks of drivers who had better performance at baseline, who tended to come from the same county as the owners (an analogue of co-ethnicity in the Liberian context). For high-initial-performance drivers who received the monitors, their performance on non-monitored tasks deteriorated. It appears that the monitoring had adverse effects on the performance of drivers who were otherwise intrinsically motivated. Owners plausibly sought to avoid such adverse effects by not installing them for many drivers from the same county.
Ethnic divisions within firms appear to matter for performance in other ways as well. Hjort (2014) looks at how the ethnic composition of teams affects output in a flower firm in Kenya. Ethnically homogeneous teams are more productive than heterogeneous ones, and this tendency is exacerbated during a period of ethnic strife in Kenya. The impact on firm productivity is substantial. Hjort argues that the patterns are consistent with a model of taste-based discrimination against non-co-ethnics. The extent to which firms are able to mitigate such conflicts can be thought of as a component of firm capability.
Macchiavello et al. (2015) make a related point regarding gender in the context of an experiment in Bangladeshi garment factories, where most line workers are female and most supervisors are male. Both male and female employees believe, incorrectly, that female supervisors have less technical knowledge. This incorrect belief fades with exposure to female supervisors (who are randomized across production lines in their experiment). But there is a cost of overcoming the prejudices of employees, and it is not clear that it is profit-maximizing for an individual firm to to pay the cost of shifting the norm.
Returning to the question of why family ownership is so prevalent, a number of authors have argued that family control is in part a response to agency issues within firms, in particular to the problems that owners may have in inducing the behavior they desire from non-family managers. Ilias (2006) focuses on the surgical goods industry in Sialkot, Pakistan, and argues that the tendency of non-family managers to move to other firms and take clients and production knowledge with them leads families to favor family members as managers. One symptom of this behavior is that founders of firms who have more brothers end up with larger firms.[5] Cai et al. (2013) present evidence from Chinese firms that family members who are managers are paid more but have lower-powered incentives than nonfamily-member managers, consistent with the idea that family members are trusted more to act in the interests of the firm. These findings do not contradict the findings above that continued family control after the founder dies is bad for performance. But they do suggest that there is a reason why family control persists. Like piece rates in the soccer-ball example, family control may be another instance of a solution to agency problems that is initially beneficial (in the sense of reducing malfeasance under the founder) but that outlives its usefulness (once the founder dies).
3.3.4 Learning
For a given level of entrepreneurial ability and degree of resolution of agency problems, a firm’s accumulation of know-how - learning - can drive upgrading. But in many cases, know-how cannot simply be purchased on an open market or downloaded from the internet. Much of the knowledge needed to produce successfully is tacit (i.e. not written down in instruction manuals) an idea that goes back at least to Katz (1984) and Pack and Westphal (1986). In addition, many organizational capabilities need to be worked out in the practice of producing; as Gibbons (2010) puts it, they need to be “homegrown.” (See also Dessein and Prat (2019).) Learning is likely to require investments with uncertain payoffs, and to take time. This subsection reviews recent work on a number of channels through which learning can occur.
3.3.4.1 Learning within firms
An important distinction in the learning literature is between learning from one’s own experiences (i.e. learning by doing) and learning from others. There is extensive evidence from industrialized countries that firms learn by doing and that the rate of learning can vary widely across firms (see e.g. Argote and Epple (1990), Irwin and Klenow (1994), Benkard (2000), Thompson (2001), Levitt et al. (2013), and Hendel and Spiegel (2014)). To date, there has been relatively little research on specific mechanisms of learning-by-doing within larger manufacturing firms in developing countries. One exception is the recent study by Menzel (2019), which uses detailed production data from three multi-floor garment factories in Bangladesh and finds that knowledge about how to produce new designs spills over across production lines on the same floor (which correspond to organizational subdivisions of the companies), but not across floors. Atkin et al. (2017b), discussed above, also documented a form of learning within firms.
Another form of within-firm learning is the transfer of knowledge or technologies across establishments (or across firms within a corporate group). These transfers are easier to observe when they cross international borders. Using data on foreign affiliates of US multinational firms in a large set of countries (including many developing countries), Branstetter et al. (2006) show that when countries strengthen their intellectual property protections, royalty payments for technology transferred to affiliates in those countries increase. There is also evidence for technology transfers across firms within developing countries. For instance, Jiang et al. (2018) look at innovation outcomes in international joint ventures in China, and also for firms that participate in the joint ventures (separate from the joint venture themselves), and find that such partner firms see within-firm increases in patenting rates following the establishment of the joint venture. (See also Bai et al. (2019).)
3.3.4.2 Learning from other firms
Besides learning from their own experiences, firms also clearly learn from others. Although perhaps the strongest evidence of such learning spillovers comes from developed countries (Irwin and Klenow, 1994) or agriculture in developing countries (Foster and Rosenzweig, 1995; Conley and Udry, 2010), there is also growing evidence that manufacturing firms in developing countries learn from other firms. The learning spillovers may occur through suppliers, buyers, peers, or workers, among other channels.
Learning from suppliers was discussed briefly above in the context of the FDI spillovers literature. There is also evidence of learning through suppliers shared with foreign firms. Using a survey of Bangladeshi garment firms that elicited the top three suppliers of each firm, Kee (2015) finds that local “siblings” of foreign-owned firms, which share a local supplier, increased productivity and product scope when for arguably exogenous reasons the market share of the foreign-owned sibling expanded. Although these effects could simply reflect greater availability of particular types of inputs, Kee suggests that the most important channel is knowledge flows. As noted above, Fieler et al. (2018) argue that quality upgrading by some producers can lead to quality upgrading by nearby firms that share suppliers.
Studies on learning from selling to foreign buyers or to locally based multinationals were discussed in Section 3.1 above. To date, it appears that there have been few studies in developing countries of learning from buyers who are not multinationals or on the export market. Evaluating the magnitude of spillovers from domestic buyers versus international buyers, and how these relate to product quality, seems to be a promising area for research.
Learning spillovers from peers, widely believed to exist, are challenging to document empirically, in part because of thorny econometric problems in estimating social effects (Manski, 1993). But recent studies have been able to manipulate experimentally the peer groups of entrepreneurs, to gain leverage for econometric identification. In an important contribution, Cai and Szeidl (2017) randomly assigned managers from 2,820 Chinese firms into groups that met monthly for one year. The meetings had a large effect on firm revenues (8.1%) and also had positive effects on profits and a management practice index similar to the World Management Survey score. To explore the learning channel directly, the authors randomly allocated information about a government grant and a high-return savings opportunity for managers, and found that not-directly-informed managers in groups where others had received the information were more likely to apply for both programs than not-directly-informed managers in groups where others had not received the information. In addition, they find that information about the government grant, which was plausibly perceived as more rival than the savings opportunity, was less likely to spill over when more firms in the group were direct competitors. No such difference is evident for the manager savings opportunity, which was less rival. Together, the results provide compelling evidence of learning spillovers between firms.
The Cai and Szeidl (2017) results contrast somewhat with a similar, earlier intervention by Fafchamps and Quinn (2018). By randomly assigning local entrepreneurs as judges in business-plan competitions in Ethiopia, Tanzania, and Zambia, Fafchamps and Quinn successfully generated in experimental variation in the judges’ peer networks. But the effects overall were quite modest. The authors found no effects on diffusion of management practices, client and supplier relations, or innovation, although they did find effects on tax registration and having a bank account (correcting for multiple hypothesis testing). The contrast with the Cai and Szeidl (2017) study is likely due in part to differences in the intensity of the peer interactions (in Fafchamps and Quinn (2018), the entrepreneurs met only once, rather than monthly for a year as in Cai and Szeidl (2017)) and in part to sample size (345 entrepreneurs in Fafchamps and Quinn (2018), 2,820 in Cai and Szeidl (2017)).
Two other notable recent studies have explored learning from peer firms in an experimental or quasi-experimental setting. Hardy and McCasland (2016) randomly allocated a new technology for weaving garments and training in using the technology and and they experimentally generated demand for products that required the technology. As in Cai and Szeidl (2017), they find that entrepreneurs are more likely to share information when they face less head-to-head competition. Although not focused on developing-country firms, Giorcelli (2019) is one of the few studies able to examine longterm outcomes of exposure to other firms. Under the Marshall plan in the 1950s, the US government sponsored trips of Italian managers to US firms and subsidized purchases by the Italian firms of advanced US technology. Giorcelli compares the set of firms that participated in the program to a set of firms that applied and were accepted but because of subsequent budget cuts were not able to participate. The sales, employment, and productivity of firms that participated in the trips rose quickly and continued to rise steadily for at least 15 years. The productivity of firms that only received the technology subsidies also rose but reached a plateau after ten years. Outcomes for firms that received both were significantly greater than the sum of the effects for each alone, suggesting that there were complementarities between the trips and the technology subsidies.
Another channel through which firms may learn from other firms is employee flows. In one famous example, employees of a single Bangladeshi garment firm, Desh Garment Company, a joint venture with Daewoo Corporation, were sent to Korea for training in production techniques. More than 100 Korea-trained Desh employees subsequently moved to new firms, in many cases starting their own firms. These flows were an important catalyst for the growth of the Bangladeshi garment sector (Rhee, 1990; Rhee and Belot, 1990; Mostafa and Klepper, 2018). Recent papers have provided evidence on several types of spillovers through worker flows, although not (for the most part) on upgrading outcomes. Using Brazilian employer-employee data, Poole (2013) finds that when Brazilian firms hire workers who have previously worked in an MNC, the wages of incumbent workers rise.[6] Researchers have also found evidence that employee movements lead “receiving” firms to export to similar destinations (e.g. Mion and Opromolla (2014) and Mion et al. (2016) in Portugal) and import from similar origins (e.g. Bisztray et al. (2018) in Hungary) as “sending” firms. Econometric identification of spillovers is always a challenge, but the accumulation of consistent findings raises one’s confidence that worker flows are an important channel for knowledge flows.
3.3.4.3 Learning from trainers/consultants
In addition to learning from their own experiences and learning from other enterprises, firms can also learn from trainers and consultants, whether their services are subsidized by governments or NGOs or purchased at market prices. An influential review of training experiments by McKenzie and Woodruff (2013), focused on small and medium-sized enterprises (SMEs), finds that most studies have very wide confidence intervals, with the result that it is rarely possible to reject a null hypothesis of no impact.[7] (See also the reviews by Grimm and Paffhausen (2015) and Quinn and Woodruff (forthcoming). Because the literature has been thoroughly discussed in these previous reviews, here I will primarily focus on a few contributions that seem particularly relevant.
Bruhn et al. (2018) randomly allocated heavily subsidized consulting services, provided by private consulting firms, to SMEs (average employment: 14) in Puebla, Mexico. The intervention was of moderately high intensity: the firms met one-on-one with consultants for four hours per week for one year. There was not a uniform body of advice given; the consultants tailored their messages to the needs of the individual firms. The authors estimate positive short-term effects on productivity and return on assets, although these effects are only marginally significant (at the 10% level) and not robust in all specifications. By linking the experimental sample to administrative data from the Mexican social security agency, the authors were able to document significant effects on employment over a longer term (5 years).
Perhaps the most influential contribution in this area has been the consulting experiment of Bloom et al. (2013) in 17 Indian textile firms. The intervention was intensive: it provided one month of consulting from a multinational consulting firm to both treatment and control firms (the “diagnostic phase”) and then four months of consulting to treatment firms only (the “implementation phase”). The market value of the consulting services for the treated plants was approximately $250,000 USD per firm. The authors tracked 38 specific management practices, including performing regular maintenance on machines, tracking inventories at least weekly, monitoring quality defects daily, and offering performance pay to non-managerial and managerial staff. Using several methods to address concerns about small sample size, the authors find clear evidence that the implementation-phase consulting was effective both in increasing the share of the 38 management practices that firms adopted and in improving firm performance, measured in terms of output, TFP, or reductions of quality defects and inventory. The authors also use the consulting treatment as an instrument for the share of the 38 management practices adopted, to estimate the effect of the practices on performance (output, TFP, quality, inventory) and find significant coefficients on the management-practices variable. In a followup paper, Bloom et al. (forthcoming) find that the effects were still present nine years later: firms treated in the original experiment continued to employ more of the management practices, had greater worker productivity and higher-quality looms, and were more likely to be exporters.
This project has broken significant new ground in the study of firm behavior, and has rightfully been influential. But three notes of caution are in order. First, to interpret the instrumental-variables (IV) results as evidence for a causal effect of the specified management practices requires the exclusion restriction that the consulting affected performance only through its effect on the share of the 38 management practices adopted. If one believes that the four months of intensive consulting had effects on firm behavior that are not captured by the share-of-the-38-practices variable, then one should not interpret the IV estimates as causal effects of the management practices themselves. For this reason, this study should arguably not be considered definitive evidence for the “vertical” view, discussed in Section 2.2.3 above, that the 38 practices (or some subset of them) are better than existing practices across contexts.[8] Note that this exclusion-restriction concern does not apply to the reduced-form (Intent-to-Treat) estimates of the effects of the consulting itself on performance, which are compelling. Second, the returns to the intervention are imprecisely measured. The authors did not have access to internal accounting data from the firms, and instead estimated profits based on their own performance estimates and a series of assumptions about the cost savings from reduction of waste fabric, profits expected to be derived from increased output, and other factors. On the basis of these assumptions, they estimate a return of $325,000 USD per year on the $250,000 USD worth of consulting services. Estimating profits in this way is an inexact science, and there is likely to be both significant heterogeneity and significant ex ante uncertainty in the profit effects.[9] Third, relatedly, it is not clear that firms were making mistakes by not adopting the management practices on their own. Although the authors themselves are careful to attribute the lack of adoption to a lack of information, the paper appears to have been interpreted by others as showing that firms left money on the table, since the management practices themselves were cheap to implement (about $3,000 USD). But if we interpret the cost of consulting as part of the cost of adopting the new management practices, and allow for heterogeneity and uncertainty in the returns, then it is not obvious that firms left money on the table.[10]
The Bloom et al. (2013) intervention was expensive, and it is worth investigating whether similar outcomes can be achieved more cheaply. Partnering with the Colombian government and focusing on autoparts firms, Iacovone et al. (2019) do this by comparing an intervention involving one-on-one consulting provided by local consultants (as opposed to more-expensive international consultants) to an intervention involving group consulting. The aim of the group consulting was to reduce costs and to take advantage of firms learning from one another. The authors find that both interventions had an effect on management practices, and that the group-consulting intervention (but not the individual consulting) had positive effects on employment and sales. Neither intervention had a significant positive effect on productivity, although the confidence bands are wide. Given that the group-consulting intervention is less costly, the study suggests that it would be the preferable design for scaling up.
The literature on training and consulting interventions is growing quickly. Several notable recent papers find positive effects on firm performance. Higuchi et al. (2017) randomized classroom and onsite training to 312 small manufacturers in Vietnam (average employment: 20), tracking firms over five years, and find positive effects on survival, sales, value-added, and profit. Higuchi et al. (2019) randomized classroom and on-site training, including quality control and production management practices as well as more standard topics such as marketing and record-keeping, to 113 small garment manufacturers in Tanzania (average employment: 5) and find positive effects on sales, value-added, and the number of products sold after 3 years. (See also Higuchi et al. (forthcoming).) Anderson et al. (2018) randomized marketing and financial skills training across 852 small enterprises in South Africa (average employment: 2.4), and find positive effects on profits, sales, and employment among the marketing group and on profits and cost-reductions among the financial-skills group.
Overall, although several studies have documented positive impacts, the effects of training and consulting interventions appear to be sensitive to the content of the advice and the details of implementation. The most successful interventions have tailored advice to the particular needs of firms, rather than providing cookie-cutter guidelines. It has often been important to follow firms over several years to see significant effects. The most successful interventions have been intensive, and in several cases expensive. Questions remain about whether firms leave money on the table by not purchasing training or consulting services and about which approaches are most cost-effective. At the same time, it seems clear that training and consulting can have significant positive effects on firm performance.
[1] The recent review by Kremer et al. (2019) devotes a section to “behavioral firms” but asserts that “we have a limited understanding of what the objectives of firm-owners in developing countries are” (p. 418).
[2] Relatedly, Adhvaryu et al. (2019b) find that more attentive managers are more effective in reallocating workers in response to negative worker-level productivity shocks from pollution exposure.
[3] See e.g. the reviews by Gibbons (2010), Gibbons and Henderson (2013), Lazear and Oyer (2013), and Garicano and Rayo (2016). Bandiera et al. (2011) review related work on how social connections and incentives can affect productivity.
[4] The cost reduction is modest, approximately 1% of total costs, but the fixed costs of adoption are also modest. The authors calculate the time required to recoup the fixed costs to be less than 8 weeks for 75% of firms in the treatment group.
[5] Bloom et al. (2013) make a similar observation about the Indian textile firms they study.
[6] See also Stoyanov and Zubanov (2012) and Labanca et al. (2014).
[7] Strikingly, in two interventions with tailors in Ghana, the impact on profits dipped negative before firms reverted to their previous practices (Karlan et al., 2015).
[8] The Atkin et al. (2017b) soccer-ball study provides one example where performance pay (in the form of piece rates) got in the way of technology adoption, and a less high-powered incentive scheme appeared to be more conducive to learning. See also Verhoogen (2016).
[9] The follow-up paper, Bloom et al. (forthcoming), was unfortunately unable to measure profits or productivity.
[10] In the notation of Section 2.1, the costs of acquiring knowledge and capabilities (IJ, IK, and Iλ) may be sufficiently large that it is not worthwhile for the firm to incur them, given the heterogeneous and uncertain benefits. Recent work by Alfaro-Serrano (2019) emphasizes these costs of adoption and shows that a Peruvian program to subsidize certifications such as ISO 9001, which require formalization and documentation of processes but not particular management practices, had the indirect effect of increasing adoption of higher-scoring management practices.
