Market Power and Monetary Policy Speech given by Andrew G Haldane Chief Economist Bank of England Co-authors: Tommaso Aquilante, Shiv Chowla, Nikola Dacic, Riccardo Masolo, Patrick Schneider, Martin Seneca and Srdan Tatomir. Federal Reserve Bank of Kansas City Economic Policy Symposium Jackson Hole, Wyoming 24 August 2018 The views expressed here are not necessarily those of the Bank of England or the Monetary Policy Committee. I would like to thank Federico Di Pace, Laure Fauchet, Rebecca Freeman, Jeremy Leake, Clare Macallan, Colm Manning, Roland Meeks, Kate Reinold, Natalja Sekhan, Silvana Tenreyro, Jan Vlieghe and Robert Zymek for their comments and contributions. This work contains statistical data from ONS which is Crown Copyright. The use of the ONS statistical data in this work does not imply the endorsement of the ONS in relation to the interpretation or analysis of the statistical data. This work uses research datasets which may not exactly reproduce National Statistics aggregates. 1 All speeches are available online at www.bankofengland.co.uk/speeches The topic for this panel is the link between developments in product markets and monetary policy. It is a great one. A lot of attention has been paid by central bankers over recent years to the relationship between labour markets and monetary policy (for example, Yellen (2014) and Constâncio (2017)). And rightly so. The relationship between monetary policy and product markets has, by comparison, been the road less travelled.1,2 Labour markets have been subject to big structural shifts over recent years, including the secular fall in the degree of worker unionisation in a number of industries (for example, Schnabel (2013)), the emergence of the so-called “gig economy” (for example, Taylor (2017) and Katz and Krueger (2017)) and secular rise in the degree of globalisation and automation in the workplace (for example, Brynjolfsson and McAfee (2014) and Acemoglu and Restrepo (2018)). Each of these shifts has led to a change in employment patterns and tenures and in workers’ bargaining power. These structural shifts have been used to help explain the secular fall in labour’s share of national income and the recent weakness of wage growth across a number of advanced economies (for example, Dao et al (2017) and Abdih and Danninger (2017)).3 They have also been used to justify potential shifts in the position and/or the slope of the Phillips curve (for example, Blanchard (2016) and Kuttner and Robinson (2010)). Each of these potentially has a bearing on the setting of monetary policy. Yet, over the same period, structural shifts in the product market have been no less profound. They include the emergence of highly-integrated global supply chains, increasing the degree of specialisation of product markets (Baldwin (2016)); the blossoming of companies benefitting from global network economies of scale and scope, who acquire “superstar” status (Autor et al (2017)); and the rapid emergence of e-commerce and price-comparison technology (Cavallo (2017)). The associated shifts in market power, too, might plausibly have altered some of the key macro-economic relationships in the economy (De Loecker and Eeckhout (2017)). They may have influenced the pricing and provision of goods and services in the economy and hence the Phillips curve. And they may have influenced the amount of investment and innovation undertaken by firms and hence the aggregate demand curve (Aghion et al (2005)). They, too, might thus have a bearing on the setting of monetary policy. These structural shifts in product and labour markets may, in some cases, have had common cause. For example, network economies of scale and scope could potentially have increased some companies’ market power both over their labour inputs (through monopsony effects) and product outputs (through monopoly effects). This could show up in both a falling labour and a rising profit share, with potential macro-economic implications for activity, costs and prices (Autor et al (2017), Barkai (2017)). 1 As Blanchard (2008) said, “How mark-ups move, in response to what, and why, is however nearly terra incognita for macro.” 2 Some notable papers that discuss the impact of product market developments on the macro-economy include Cacciatore and Fiori (2016) and Eggertsson, Ferrero and Raffo (2014). 3 Unlike many other advanced economies, it is worth noting that the UK labour share has not been on a downward trend. 2 All speeches are available online at www.bankofengland.co.uk/speeches 2 To explore these issues, we start by discussing briefly recent empirical evidence on market power and its potential macro-economic explanations and implications. We then explore its effects on monetary policy, using counter-factual policy simulations and adaptations of a simple New Keynesian model. Taken together, this evidence suggests that an increase in market power and mark-ups could have potentially important consequences for the economy and policy. These are summarised, in stylised terms, in Figure 1. To the extent a secular rise in mark-ups reflects a set of trade-off inducing shocks, that would shift outwards the output/inflation variability (policy possibility) frontier (from A to B). It may steepen the Phillips curve, causing the policy possibility frontier to rotate clockwise (B to C). And it may also potentially alter the optimal weights placed on output and inflation stabilisation by the policymaker, shifting the point of tangency between the policy possibility frontier and policymakers’ loss function (C to D). Taken together, the net effect of increased market power could be a potentially significant rise in inflation (but less so output) variability, relative to the counterfactual case of stable and static mark-ups (A to D). As for monetary policy, the fact that these are trade-off inducing shock places limits on its stabilisation capacity. The (level and variability) of the optimal interest rate path is, as a result, less affected by increased market power, despite significant shifts in policy possibility frontiers and policymaker preferences. There are two ways in which the path of monetary policy might potentially be affected to a greater degree by increases in market power. When companies have a significant degree of market power, the level of output produced is likely to be below the social optimum, creating an incentive for monetary policy to try to offset that by running the economy hotter (“inflation bias”). And if market power lowered companies’ investment rates, this could reduce the economy’s neutral rate of interest. Neither, however, at present has a strong empirical basis. To the extent these channels do operate, they reinforce the institutional case for independent central banks charged with pursuing well-defined inflation targets. At the same time, actual inflation across advanced economies has of course been relatively low and stable over recent decades. So while the micro-economic evidence – a secular increase in mark-ups – is striking, it is not easily reconciled with the macro-economic evidence on measured inflation, including on the impact of mark-ups on inflation (for example, Smets and Wouters (2007)). Reconciliation of the two strands of evidence – micro and macro – means that some combination of the following would have to be true. First, the micro-economic firm-level evidence may not accurately describe how economy-wide mark-ups have evolved since the 1980s. Second, other macro-economic factors may have more than offset the impact of rising mark-ups on the behaviour of inflation. Third, the theoretical macro-economic framework we use here – a New Keynesian model with monopolistic competition – may not be appropriate to analyse firm-level changes in mark-ups. The apparent puzzle between the micro-economic and macro-economic evidence deserves further research, given its potential impact on inflation dynamics and monetary policy. 3 All speeches are available online at www.bankofengland.co.uk/speeches 3 Market Power – Evidence and Implications There is a rich micro-economic literature that assesses the impact of market power on pricing and other firm decisions (for example, Tirole (1988)). There has been rather less evidence linking the industrial organisation of firms to developments in the wider macro-economy. That has changed recently, with a number of papers exploring the empirical evolution of (firm, sectoral and national) measures of market power and their implications for the macro-economy (for example, De Loecker and Eeckhout (2017, 2018) and Díez et al (2018)). Perhaps the simplest way of capturing market power is through measures of market concentration, such as Herfindahl-Hirschman Indices (HHI) (Hirschman (1964)) or concentration ratios (the share of sales that accrues to the largest firms within an economy or sector). Evidence suggests that market concentration, measured either through HHIs or concentration ratios, may have increased in the US over recent decades, across a broad range of sectors (for example, Autor et al (2017)). This pattern is not uniform, however, with concentration among European companies showing no such trend (Gutiérrez and Philippon (2018)). The evidence on industry concentration in the UK suggests it occupies a mid-Atlantic position. Chart 1 plots the turnover share of the largest 100 UK businesses since 1998 (i.e. concentration ratio).4 This ticks up in the lead-up to the financial crisis, although this pick-up is more modest than in the US, from 20% to around 28%.5 Concentration has flattened-off in the period since the crisis, however, in line with other European countries. Turning to measures of concentration within the financial services industry, the international pattern is somewhat more uniform. Chart 2 plots the largest five banks’ share of total banking assets in the US, euro area and the UK. Levels of banking concentration started fairly high, averaging around 30%. They drifted further upwards in the run-up to the crisis, although this drift was again fairly modest. Since the crisis, however, measures of banking concentration have flat-lined and, in the UK, have fallen slightly. Concentration indices have their limitations, though, and need not always be associated with market power. Some firms may be able to exercise market power in setting prices even without having a large share of a market if, for example, there is brand loyalty. And in a world of differentiated products, concentration measures such as HHIs or concentration ratios no longer correlate closely with market power (Bresnahan (1989)). With non-homogenous goods and non-Cournot competition, a better measure of market power is often provided by firms’ mark-ups – the ratio of their price to their marginal cost (De Loecker and Eeckhout (2018)). The larger the mark-up, the greater the degree of market power, whether at the firm, sector or national level. Mark-ups also have the benefit of being the relevant measure of market power in the workhorse models of the macro-economy used by policymakers. 4 Excluding financial services firms. 5 See, also, Bell and Tomlinson (2018). 4 All speeches are available online at www.bankofengland.co.uk/speeches 4 In that spirit, a number of recent papers have estimated measures of mark-ups based on individual company accounts data. These cover a wide range of companies, sectors, countries and time periods (for example, Díez et al (2018)). The findings from these studies are, in macro-economic terms, both quite striking and quite strikingly uniform in the broad trends they reveal. For example, De Loecker and Eeckhout (2018) have recently calculated mark-ups for around 70,000 firms across 134 countries over almost four decades.6 Since 1980, they estimate that the sales-weighted mark-up for the average firm across countries has risen by a remarkable 50 percentage points.7 Table 1 shows their mark-up measures for the G7 economies over the period. Though there is cross-country variation, average mark-ups have risen significantly in every G7 country, by between 30 and 150 percentage points. Taken at face value, the macro-economic implications of these shifts in mark-ups could be very large. The most direct and immediate impact would be on measured inflation rates. According to Table 1, mark-ups will have been adding, on average, over one percentage point each year to measured inflation rates across the G7 countries between 1980 and 2016, other things equal. As context, over the same period average G7 inflation rates have fallen by over 10 percentage points.8 A second potential macro-economic impact of higher mark-ups is on sales. Higher mark-ups will, other things equal, have pushed down on aggregate demand and generated a deadweight loss of consumer surplus (“Harberger triangle”). Baquee and Farhi (2018) estimate the size of this effect and find that eliminating mark-ups entirely would raise aggregate US total factor productivity (TFP) by as much as 35%.9 To better understand some of the drivers of higher mark-ups, it is useful to look at more granular data. Using a similar approach to De Loecker and Eeckhout (2017, 2018), we draw on data for around 3,500 unique UK-listed companies from the late-1980s to construct around 33,500 firm-year mark-up estimates.10 Using that methodology, Chart 3 plots a sales-weighted measure of mean mark-ups for UK-listed companies since 1987. It shows a striking rise, from 1.2 to around 1.6, over the period. This broadly mirrors international trends. Although they capture subtly different dimensions of market power, there is a weakly positive relationship between measures of mark-up and market concentration at the sector level (Chart 4), which is statistically 6 De Loecker and Eeckhout (2018) use a dataset that largely includes publicly-traded companies, but there are also some privately held firms. 7 Similar estimates have recently been provided by Díez et al (2018). 8 OECD data. 9 De Loecker and Eeckhout (2018) estimate that mark-ups in the US are a little larger than in the UK, so the boost to UK TFP from eliminating them would be smaller than the 35% estimate for the US in Baquee and Farhi (2018), albeit of a similar order of magnitude. 10 The methodology is explained in the Appendix. The data include around a little over 1000 firms, on average, per year. These firms account for around one-third of UK employment. Their sales are equivalent to around one-third of UK turnover and around two-thirds of UK nominal GDP. We exclude financial sector firms and, having estimated mark-ups, trim outliers, i.e. those firm-level mark-ups that are below the 1st percentile and above the 99th percentile of the firm-level mark-up distribution in a given year, 5 All speeches are available online at www.bankofengland.co.uk/speeches 5 significant at the firm level.11 The same has been found among companies in other countries (Díez et al (2018)). This gives some degree of reassurance that the rise in measured market power has been a genuine one. If we slice the mark-up data for non-financial companies on a sectoral basis, this suggests this rise has been reasonably broad-based (Chart 5). All but two of the ten sectors have seen mark-ups rise since 1987, although some are volatile. Six of the ten have seen them rise by more than 30 percentage points. Among the largest rises have been in manufacturing (70 percentage points), professional, scientific and technical (62 percentage points) and transport and storage (57 percentage points). This broadly mirrors the international evidence.12 One apparent exception is the banking sector. Chart 6 plots a measure of banks’ net interest margins (NIMs), as a proxy for mark-ups, in the UK, US and euro area since 1996. NIMs appear to have been broadly flat in these countries over recent decades. If anything, they may have fallen over the past decade. The latter is potentially the result of the low levels of official interest rates, constraining the ability of banks to lower their deposit rates in order to protect margins (for example, Claessens, Coleman and Donnelly (2017)). Another way of slicing the data is to ask how much of the rise in mark-ups is due to a compositional shift over time towards sectors whose mark-ups are already high and how much reflects a generalised rise in mark-ups within each sector. Chart 7 shows this decomposition for UK-listed firms. Compositional effects do not explain any of the rise in mark-ups in the UK; and even if we do the same exercise at the firm level, compositional shifts towards firms with high mark-ups cannot explain the rise. Rather, the rise in mark-ups appears to be reasonably generalised across sectors.13 Although relatively broadly-based across sectors, the rise in mark-ups need not necessarily be broadly based within sectors. One way of showing that is by looking at the evolution of the distribution of mark-ups over time (Chart 8). This suggests the increase in mark-ups is heavily concentrated in the upper tail of the distribution – companies whose mark-ups are in, say, the top quartile. Mark-ups among firms in this upper quartile of the distribution have, on average, increased by a remarkable 50 percentage points since 1987. By contrast, mark-ups among firms in the bottom three quartiles of the mark-up distribution have scarcely risen over the period. This distributional effect can also be seen from the large and widening gap between mean and median mark-ups (Chart 9). In 1987, this gap was 7 percentage points. By 2016, it had reached 44 percentage points. This strongly suggests that the rise in aggregate mark-ups over the past 30 years can largely be accounted for by a subset of high mark-up firms raising their mark-ups and/or market share. 11 While the positive unconditional correlation between average firm-level mark-ups and concentration at the sector level shown in Chart 4 is not statistically significant, a regression of individual firm-level mark-ups on market concentration in their sector (at the two-digit SIC level) shows a statistically significant positive relationship when we include firm and time fixed effects. 12 For example, Díez et al (2018) find that the majority of industries in the US have seen mark-ups rise since 1980. 13 Díez et al (2018) find that the increase in US mark-ups since 1980 is also relatively broad-based across sectors. 6 All speeches are available online at www.bankofengland.co.uk/speeches 6 This fattening of the upper tail of the mark-up distribution is not uniform across sectors. Chart 10 plots a measure of the skew of the mark-up distribution across different sectors over time. The fattening of the upper tail of the distribution is most pronounced in the ICT, transport and storage and manufacturing sectors, each of which is associated with higher average levels of mark-up. In understanding the characteristics of these firms, one revealing cut comes from taking into the account the extent to which UK-based firms’ sales are domestic or foreign-focussed (Chart 11). While both categories have seen their mark-ups rise somewhat, this has been far larger among firms selling predominantly into foreign markets (almost 60 percentage points) than domestic markets (around 15 percentage points).14 Within that, this rise in mark-ups among foreign relative to domestic sales-focussed firms is largest in the manufacturing and ICT sectors. Taken together, this evidence is consistent with a story of rising mark-ups being concentrated among internationally-operating firms, who perhaps benefit disproportionately from global network economies of scale and scope. These firms tend to occupy the fat and fattening upper tail of the mark-up distribution. These are firms that might legitimately be termed “superstars” (Autor et al (2017)). Given this diagnosis, what impact might the rise in mark-ups have had on the macro-economy? One aspect is what impact increased market power may have had on firms’ incentives to invest and innovate and hence on firms’ productivity. With investment and productivity each having under-performed over recent years, the relationship with market power has been subject to increased academic scrutiny recently (for example, Eggertsson, Robbins and Wold (2018)). The relationship between mark-ups and productivity is vital in understanding their macro-economic effects (Van Reenen (2018)). On the one hand, if highly productive ‘superstar’ firms, benefitting from network economies of scale and scope, have become more dominant, then higher mark-ups could be the side-effect of a positive supply shock in the economy. On the other hand, if mark-ups are the counterpart to increased market power and reduced competitive pressures, that would suggest a negative supply shock. These effects are not mutually exclusive. For example, Aghion et al (2005) develop a model which generates a concave relationship between competition and investment. Within some range, increased market power raises rents and acts as a spur to investment, innovation and productivity. But beyond a point, those forces go into reverse. Market power is associated with a fall in innovation and investment incentives, with knock-on negative effects for productivity. There is some empirical support for such a relationship. Jones and Philippon (2016) and Gutiérrez and Philippon (2017) suggest increased market power may have reduced investment among US companies. 14 Our results are consistent with De Loecker and Warzynski (2012) who find that exporters charge, on average, higher mark-ups and that firm mark-ups increase upon export entry. 7 All speeches are available online at www.bankofengland.co.uk/speeches 7 De Loecker and Eeckhout (2017) document a negative relationship between mark-ups and the capital share among global companies. And Díez et al (2018) identify empirically a concave relationship between mark-ups and investment, in line with Aghion et al (2005). We can re-run the Díez et al (2018) investment equations using the panel of UK-listed firms. This also finds a concave relationship with mark-ups (Table 2, column 1). The same relationship holds between mark-ups and R&D expenditure (Table 2, column 2). Chart 12 plots the estimated investment curve. It suggests that firms with mark-ups above around two tend to be associated with lower investment rates, in line with Díez et al. With estimated firm-level mark-ups having risen secularly in a number of countries, this is potentially a cause for concern. It is important, however, not to overstate the likely impact of this rise in mark-ups on aggregate investment, innovation and productivity. The rise in mean mark-ups in the UK over the past 30 years would still leave them below the levels at which investment rates start falling. The same is true among global firms. Indeed, among our panel of UK-listed companies, the shift in average UK mark-ups since the late-1980s would, using the estimated investment equation, be expected to have raised average investment rates by around 1 percentage point. Finally, there is the question of whether any potential negative effects of increased market power on investment and R&D translate into a negative effect on productivity. Evidence suggests there could be an effect. De Loecker and Eeckhout (2017) argue that, once account is taken of the rise in mark-ups, it is possible to account for the slowdown in US productivity growth after 1980. And Díez et al (2018) find that the greater the distance to the technological frontier, the lower a firm’s investment – a “reverse catch-up” effect. If we look at the relationship between productivity and mark-ups across UK-listed firms, there is evidence of a positive relationship with TFP but no significant relationship with labour productivity (Table 2, columns 3 and 4). If anything the relationship with TFP may be convex, with higher mark-up firms being associated with proportionately higher levels of total factor productivity. There is some evidence of “reverse catch-up” effects, but only at high levels of mark-ups. Overall, then, while the theoretical and empirical evidence suggests it is possible higher market power and mark-ups may have come at some cost in lower investment and innovation, the evidence is not overwhelming and certainly would not imply that the aggregate effect is large. A second relationship explored recently is between market power and the labour share.15 Autor et al (2017) find a negative empirical relationship using measures of market concentration among US companies. And Díez et al (2018) and De Loecker and Eeckhout (2018) identify a weakly negative relationship between 15 The relationship between labour, capital and profit shares is discussed in Barkai (2017). 8 All speeches are available online at www.bankofengland.co.uk/speeches 8 mark-ups and the labour share. If we run regressions similar to those in Díez et al for UK-listed companies (Table 3), we also find a negative relationship between mark-ups and the labour share.16 The analysis presented here, and much of the recent literature, is based around estimates that suggest a secular rise in firm-level mark-ups. Some caution is advisable when drawing conclusions from these results. First, some macro-economic evidence points to falling, rather than rising, company mark-ups/margins (for example, Chen, Imbs and Scott (2009, 2004)). Second, some mismeasurement may be at play in the estimation of mark-ups (for example, Traina (2018)). These uncertainties in the measurement of mark-ups should be borne in mind in interpreting what follows. Market Power and the Macro-Economy – A Simulation Approach Having assessed some evidence on the evolution, and macro-economic implications, of increased mark-ups and market power, the next question is what implications these may have for the setting of monetary policy. This does not appear to have been an extensively examined area of research, whether among academics or policymakers.17 What follows is an initial exploration of some of the potential channels. One simple way of beginning to gauge how a rise in mark-ups might affect the economy and monetary policy is to simulate their impact using a macro-economic model. For this purpose, we model the economy using the Bank of England’s in-house DSGE model, COMPASS.18 Monetary policy is assumed to follow a simple Taylor rule, with interest rate smoothing.19 The simulations are shown for a variety of different values of the relative weight policymakers place on output and inflation deviations from target in the Taylor rule. Chart 13 considers the impact on inflation, the output gap and monetary policy of a temporary mark-up shock that delivers a one percentage point increase in annual inflation. The dynamics of the economy are largely as we would expect following an adverse supply shock. The inflation rate rises, and real GDP usually contracts, in both cases temporarily. Although temporary, these disturbances are often material and always persistent, despite monetary policy acting to damp these fluctuations. The reason monetary policy struggles to damp these fluctuations is because a mark-up shock is trade-off inducing. Monetary policy is caught between loosening to return output to potential and tightening to return inflation to target. Which wins out depends, crucially, on the relative weight placed on these twin objectives in the policy rule. When inflation deviations are given greatest weight, monetary policy tightens materially. When output deviations are given greatest weight, monetary policy scarcely tightens at all. 16 Unlike Díez et al (2018), we use the reported data on staff costs in our firm-level dataset when calculating the labour share. 17 A recent exception would be the work of Mongey (2018). 18 Burgess et al (2013). 19 Taylor (1993). 9 All speeches are available online at www.bankofengland.co.uk/speeches 9 With monetary policy facing this trade-off, it follows that an increased prevalence of mark-up shocks would leave policymakers somewhat constrained in their ability to smooth the economy. Put differently, a sequence of trade-off inducing mark-up shocks would tend to worsen the trade-off between output and inflation variability, for a given monetary policy rule. The “Taylor curve” frontier of policy possibilities would be expected to shift outwards.20 To illustrate that, we can conduct a counterfactual simulation of the effects of mark-up shocks on the course of output, inflation and interest rate variability. Chart 14 shows the variability of inflation and the output gap (the black dot) generated by the model. It also shows the variability of output and inflation when the economy is re-simulated having “switched-off” the shocks to firm mark-ups identified by the model (the red symbols). Policy is again assumed to follow a Taylor rule, with varying weights on output and inflation. Mark-up shocks have a material impact on output and especially inflation variability, even with monetary policy cushioning their effects. The variance of inflation is reduced by around a quarter, and variance of the output gap by around 10%, when mark-ups shocks are switched-off. The policy possibility frontier of output/inflation variabilities is shifted outwards materially by the presence of mark-up shocks.21 The scope for monetary policy to cushion these shocks is relatively limited. Chart 15 plots the variability of interest rates alongside output variability, for the same set of policy rules. Mark-ups shocks affect interest rate variability relatively modestly.22 And the variance of interest rates is reduced by only around 5% when mark-up shocks are switched-off. This tells us that trade-off inducing shocks to mark-ups leave the (path and variability) of interest rates less affected than inflation. Clearly, this simulation places an upper bound on this shift as it effectively removes shocks to mark-ups. In practice, the evidence on how the variability of mark-ups may have evolved is mixed. On the one hand, macro evidence suggests a fall in the variability of both output and inflation in many countries recently, a finding that has been attributed by some to a lower incidence of mark-up shocks (for example, Smets and Wouters (2007) and Kapetanios et al (2017)). On the other, direct micro-level evidence on mark-up behaviour over recent years suggests a potential pick-up in their trend and variability. To the extent that the historical evidence is consistent with a sequence of larger mark-up shocks, it would be expected to have made the task of monetary policymakers somewhat harder. Both output and inflation will have deviated more significantly and persistently from their long-run values. And although interest rates will have been adjusted somewhat more often in response, the trade-off inducing nature of these shocks places constraints on the degree of stabilisation monetary policy can achieve. 20 Taylor (1979). 21 The larger proportionate reduction in inflation than in output gap variability arises because mark-up shocks account for a larger proportion of the historical variance of inflation than output. 22 The interest rate smoothing term in the COMPASS Taylor rule will also have a role to play here in limiting the extent of the policy response to the mark-up shock. 10 All speeches are available online at www.bankofengland.co.uk/speeches 10