Why Does Aggregate Earnings Shocks Predict Future Inflation Shocks? *

Transcription

1 Why Does Aggregate Earnings Shocks Predict Future Inflation Shocks? * Lakshmanan Shivakumar London Business School lshivakumar@london.edu Oktay Urcan University of Illinois at Urbana Champaign ourcan@illinois.edu October 20, 2014 Abstract Prior studies document that aggregate earnings news is related to news about future inflation. We propose two alternative explanations for this relation - one based on firm s changing their investments in response to profitability shocks and the other based on consumers varying their consumption in response to higher disposable income or greater wealth arising from increased profitability of listed firms. Since supply of goods and services is relatively inelastic in the short-run, our arguments imply that shocks to demand for investments (consumption) will affect prices of investment (consumption) goods and services. Consistent with aggregate earnings news affecting business investments, we find that profitability shocks predict investment shocks in subsequent quarters as well as shocks to Producer Price Index, which primarily tracks prices of production-related goods. We find no evidence that aggregate earnings surprises contain information about future consumption shocks or shocks to prices of consumption goods (Consumer Price Index). Our analyses also reveal aggregate earnings surprises predict future investment and PPI forecast-errors and that this forecasting inefficiency explains previously documented inefficiencies in GDP growth forecasts. * We appreciate helpful comments from Bill Cready, Yaniv Konchitchki, S.P. Kothari, Peter Pope and seminar participants at Harvard Business School, Cass Business School and the 11 th London Business School Accounting Symposium. Oktay Urcan acknowledges the financial support from London Business School Research and Materials Development (RAMD) Fund.

2 1. Introduction Kothari, Lewellen and Warner (2006) document a puzzling negative relation between aggregate earnings growth and stock-market returns. 1 This result appears to have rekindled academic interest in developing a better understanding of the link between corporate earnings and the macroeconomy. 2 Subsequent research has attempted to discern the precise nature of information contained in aggregate earnings growth and generally shows that aggregate earnings reflect information about future inflation. For instance, Shivakumar (2007) finds that aggregate earnings growth is positively associated with levels of future inflation. Patatoukas (2014) documents that aggregate earnings changes are positively associated with contemporaneous quarterly changes in expected inflation. Also, consistent with aggregate earnings providing inflation-related news to investors, Cready and Gurun (2010) document a negative relation between both stock- and bondmarket returns and contemporaneous earnings information measured over short-event windows. Despite these recent developments linking aggregate earnings surprises to news about future inflation, little is known about why aggregate earnings predicts future inflation levels and shocks. While existing studies document a clear association between aggregate earnings and future inflation, they do not establish causality. A proper understanding of the mechanism through which aggregate earnings news affects future inflation is important to attribute causal links between these variables. Further, responses of monetary authorities to profitability shocks also depend upon how profitability shocks propagate in the economy. Finally, the channel through which corporate profitability shocks affect future macroeconomic variables also has implications for forecasts issued by macro-forecasters. In this paper, we offer and test two potential explanations for this predictive ability one based on demand for investment products and the other based on demand for consumption goods. Investment demand hypothesis suggests that aggregate earnings causes a shift in the demand for investments in inventory of goods held for resale as well as for investments in goods and services 1 Throughout, we refer to earnings aggregated across listed companies as aggregate earnings. 2 Recent studies examining the link between aggregate accounting information, macroeconomy and market returns include Anilowski et al. (2007), Shivakumar (2007), Bali et al. (2008), Ball et al. (2009), Sadka and Sadka (2009), Hirshleifer et al. (2009), Cready and Gurun (2010), Basu et al. (2010), Bonsall et al. (2013), Ogneva (2013), Arif and Lee (2013), Konchitchki and Patatoukas (2014a), Konchitchki and Patatoukas (2014b), Gkougkousi (2014) He and Wu (2014) and Patatoukas (2014). 1

3 needed to build or enhance a firm s operating capacity (hereafter, we refer to goods held in inventory as well as goods and services needed for capital expenditures as production goods.) This shock to investment demand leads to an increase in prices of production goods in the short-term, as supply of production goods are relatively inelastic in the short-term. Aggregate earnings surprises can affect future investment demand for at least three reasons. First, managers could interpret positive earnings shocks as implying a favorable change of environment for the firms goods and services and respond by investing in operating capacity. Similarly, adverse earnings shocks could cause managers to delay or cancel investments. Secondly, to the extent that a firm s profits are quickly converted to cash or liquid assets, positive shocks to earnings can relax financial constraints facing a firm by making more internal funds available for investments. Thirdly, since profits are a key determinant of credit risks, banks and other capital providers could view improvements in profits as lowering credit risks and increase corporate lending, which again makes more funds available for investments by financiallyconstrained firms. Along similar lines, negative earnings news would tighten firms financial constraints, lowering their ability to invest. The Consumption demand hypothesis, on the other hand, predicts that aggregate earnings shocks cause investors and employees to change consumption, which due to inelasticity of supply of consumption goods in the short-run, causes prices of consumption goods to adjust. Such effects arise because higher profitability increases dividend payouts, increases individual agent s net wealth through increased share prices, or possibly leads to higher wages, salaries, bonuses and commissions paid to employees. To the extent that economic agents immediately consume their increased net wealth, demand for consumption goods increases. Similarly, periods of negative earnings surprises could see consumers tightening their expenditures, causing a decline in consumption goods and affecting prices of consumption goods. The two non-mutually exclusive hypotheses not only differ in the path through which aggregate earnings surprises affects subsequent short-run inflation, but also have different implications for which measures of inflation, inflation of production goods or consumption goods, are affected by aggregate earnings. Moreover, failure to find empirical support for either of the above 2

4 hypotheses increases the likelihood that aggregate earnings changes merely reflect, rather than cause, future inflation shocks. 3 We begin our empirical analyses by examining the impact of aggregate earnings shocks on future investment and consumption shocks using a vector auto-regression (VAR) framework. We find that the aggregate of all earnings surprises disseminated publicly over a three-month window is positively correlated with subsequent quarter s investment. In particular, we document that aggregate earnings shocks are positively related to future shocks in gross private domestic investment, private nonresidential fixed investment, gross fixed capital formation and aggregate inventories. However, we do not find any significant relationship between aggregate earnings surprises and future innovations in consumption related variables (i.e., personal outlays, personal consumption expenditure, personal income and disposable personal income). Moreover, robustness tests confirm the relation between aggregate earnings and investments even in firm-level analysis. These findings suggest that aggregate earnings impacts future inflation through investment demand as opposed to consumption demand. We next investigate the relation between aggregate earnings surprises and inflation shocks, using Consumer Price Index (CPI) to track changes in prices of consumption goods and Producers Price Index (PPI) to track changes in prices of production goods. Based on Vector Auto Regression (VAR) analysis, aggregate earnings surprises is found to be positively correlated with innovations in actual PPI for finished goods, PPI for intermediate goods and PPI for crude materials in the next one to two months. However, we do not find any significant relation between aggregate earnings surprises and future innovations in CPI. This indicates that aggregate earnings shocks do not affect the shortterm prices of consumption goods and is consistent with our earlier evidence that aggregate earnings positively impacts only future investment, but not personal consumption or personal income. It is worth emphasizing that the relation between PPI and aggregate earnings cannot be mechanical, as our 3 If aggregate earnings merely reflect future inflation shocks, then it raises a puzzling question of why earnings, but not other financial variables included as controls in the analyses, reflects future macro shocks. 3

5 investigations focus on the predictive ability of aggregate earnings news for future inflation shocks rather than on contemporaneous relations. 4 Although our results consistently point to aggregate earnings shocks affecting demand and prices only of production goods, evidence in extant literature of a link between aggregate earnings and future inflation is largely based on using CPI as a proxy for inflation. To reconcile this difference, we re-examine evidence from prior studies on the relation between aggregate earnings and future CPI. Consistent with Shivakumar (2007), we find that aggregate earnings news appears to be positively correlated with future CPI inflation levels. However, once we control for information in lagged CPI, the relation between aggregate earnings news and CPI turns insignificant, consistent with our VAR findings. Similarly, we show that the relation between CPI news and aggregate earnings documented in Gallo et al. (2013) are sensitive to the inclusion of financial variables as controls. Our findings suggest that aggregate earnings contain information primarily about future PPI inflation innovations that is incremental to information contained in lagged values of other macroeconomic and financial variables. These results also suggest that aggregate earnings are more likely to affect future inflation through its effect on demand for production goods rather than through increased consumption by individuals. Konchitchki and Patatoukas (2014a) document that macro-forecasters do not fully incorporate GDP-growth information in aggregate earnings. Since investment and consumption are two of the main components of GDP, our findings raise the possibility that the GDP forecast inefficiency is driven through inefficiencies in forecasting investments or PPI. We test this possibility by studying the efficiency of investment, consumption and inflation forecasts obtained either from Money Market Service surveys or Survey of Professional Forecasters provided by Federal Reserve Bank of Philadelphia. 4 An association between corporate profits earned in a quarter and inflation in the contemporaneous quarter can be mechanical through revenues reflecting current selling prices and expenses reflecting historical costs (viz. cost of goods sold reflecting prices from the time of inventory purchase). See Ball, Kothari and Watts (1993) for a similar argument. 4

6 Our analyses reveal that aggregate earnings surprises predict errors in macroeconomists forecasts of next quarter s investment and next two month s PPI. However, no consistent inefficiency is observed for consumption or CPI forecast errors. Moreover, investment forecast errors and PPI forecast errors entirely explain the ability of aggregate earnings to predict GDP growth forecast errors, confirming that GDP growth forecast inefficiency is due to macro-forecasters missing out investment and PPI information in aggregate earnings news. Overall, the various analyses consistently document evidence that aggregate earnings shocks affects future investment demand and future prices of production goods, but has little discernable effect on future demand or prices of consumption goods. The results support the hypothesis that firms adjust their investment plans in response to profitability shocks, which causes prices of production goods to be affected in the short-run when supply of these goods and services is relatively inelastic. Our tests on forecasting efficiency shows that macro-forecasters do not efficiently incorporate the information in aggregate earnings surprises about future prices of production goods. Our paper makes several contributions to the literature. As pointed out by Ogneva (2013), identifying the precise information in aggregate earnings is important to understand the role of accounting information in the macroeconomy and in shaping overall market s expectations. While the extant literature generally recognizes that accounting earnings predict future macroeconomic variables, ours is the first study to evaluate the reasons behind this predictive ability. Also, through identification of the mechanisms underlying the link between aggregate earnings and future inflation shocks, our results highlight the importance of selecting appropriate inflation proxies in studies of aggregate earnings and macroeconomic activities. The results show that using CPI or GDP-deflators as generic proxies of inflation in studies of aggregate earnings may be insufficient. Secondly, significant anecdotal evidence and academic research suggest that firm-level earnings news affects firm s real decision. We extend this literature to study the real effects of unexpected earnings at the macroeconomic level. While the macroeconomic literature focusses largely on propagation of specific-types of shocks, viz. production shocks, input price shocks, credit 5

7 shocks, technological shocks on investments and corporate profits, our focus is on understanding the role of aggregate profitability shocks, which can be viewed as a timely metric reflecting the net effect of all macro-shocks, on aggregate investment, consumption and inflation. 5,6 Finally, our study contributes to the macroeconomics and finance literatures evaluating macro-forecasting efficiency. Given the importance of accurate inflation and other macro forecasts for a variety of economic and business decisions, a vast literature evaluates the efficiency of macro forecasts and of stock market reactions to macro releases. 7 However, these studies do not consider the information about investment demand and prices of production goods reflected in aggregate earnings news. Our study shows that macro-forecasters fail to fully incorporate this information in their forecasts. Consistent with the extant literature on aggregate earnings news, the focus of the study is on predictive ability of aggregate earnings news that is incremental to information in other sources. Naturally, our analysis is limited to the short-term, since over time other information sources will reflect information in aggregate earnings news. The paper does not investigate the longer-term consequences of aggregate earnings shocks. The rest of the paper is organized as follows. In the following section, we discuss prior literature on the relation between aggregate earnings and inflation and present alternative hypotheses linking aggregate earnings and future inflation innovations. Section 3 discusses differences between CPI and PPI measures of inflation. In Section 4, we explain our sample selection procedures and 5 Macroeconomic studies of corporate profits have typically focused on how fiscal, monetary and other macro shocks impact corporate profits (e.g., Alesina et al., 2002). Similarly, although a long list of macroeconomics studies have modelled financial frictions, most of these studies have focused on the role played by the financial sector in propagating shocks that originate in other sectors, such as productivity and monetary shocks. Recent studies (e.g., Benk et al., 2005 and Jermann and Quadrini, 2012) evaluate the role of financial frictions, defined as financing constraints or liquidity shocks, in causing economic fluctuations. 6 Also, macroeconomics studies typically rely on corporate earnings statistics collected and released by the Bureau of Economic Analysis (BEA) or similar national institutions. Such data are typically released either quarterly or annually, which can be less timely for causal analysis of real effects compared to the almost daily aggregation that is possible with individual firms preliminary earnings announcements. For instance, Cready and Gurun (2010) aggregate earnings news at a daily level to study the causal relation between aggregate earnings news and capital market returns. 7 These studies include Pearce and Roley (1985), Hardouvelis (1987), McQueen and Roley (1993), Aggarwal, Mohanty and Song (1995), Balduzzi, Elton and Green (2001) and Flannery and Protopapadakis (2002). 6

8 describe the data. Section 5 presents results from empirical analysis of aggregate earnings surprises and macro innovations, while Section 6 presents analyses of macro-forecasts. Section 7 concludes. 2. Prior Literature and Hypothesis Development 2.1 Aggregate earnings and inflation Kothari et al. (2006) document a surprising negative relation between aggregate earnings changes and stock market returns. Gkougkousi (2014) documents a similar relationship in the bondmarket for investment-grade, but not high-yield, corporate bonds. 8 To explain the negative relation, Kothari et al. (2006) posit that aggregate earnings changes contain new information about future discount rates and show that aggregate earnings growth is strongly correlated with changes in several discount rate proxies, such as T-bill rates, slope of term structure and default spread. Sadka and Sadka (2009) offer an alternative explanation and suggest that, rather than reflecting new information, aggregate earnings changes are negatively correlated with expected market returns. In order to distinguish between these alternative explanations, Patatoukas (2014) decomposes stock market returns into expected returns, cash flow news and discount rate news components and studies the relation between aggregate earnings surprises and each component of stock market returns. He shows that the stock market returns are positively correlated with expected component of returns and cash flow news, but are negatively correlated with discount rate news. To get a deeper understanding of the relationship between aggregate earnings and stock returns, recent research has turned its attention to discerning the information embedded in aggregate earnings. Motivated by the fact that corporations are a critical segment of the macroeconomy, Shivakumar (2007) evaluates whether aggregate earnings changes predict a variety of future macroeconomic activities, including industrial production growth, real GDP, and inflation. He finds that aggregate earnings are related only to future inflation, measured as CPI. Using earnings announcement returns to proxy for earnings news, Cready and Gurun (2010) conclude that aggregate earnings surprises reflect news about future discount rates. Moreover, they provide evidence that 8 In contrast to the evidence based on aggregate earnings changes, Anilowski et al. (2007) document a positive association between aggregate proportion of upward or downward earnings guidance provided by management and contemporaneous stock market returns. 7

9 earnings announcement returns, aggregated across all firms announcing earnings in a 3-day-window, are positively correlated with inflation-changes reflected in treasury inflation-protected securities (TIPS), which tracks the CPI. Cready and Gurun (2010) conclude that aggregate earnings news convey inflation news to capital markets. Consistent with Cready and Gurun (2010), Patatoukas (2014) reports a positive relation between aggregate earnings changes and contemporaneous changes in expected inflation, measured as the forecasts of 1-year-ahead GDP deflator growth rate. Konchitchki and Patatoukas (2014a) document that aggregate earnings growth predicts nominal GDP growth rates in the subsequent four quarters and also predicts forecast errors to nominal GDP growth rates. Gallo et al. (2013) argue that aggregate earnings contain incrementally useful information about the economy over major leading indicators and consequently, is used by Federal Reserve in monetary policy decisions. Consistent with this argument, they show that aggregate earnings growth predicts future changes in monetary policies. Konchitchki and Patatoukas (2014b) evaluate whether financial statement analysis of aggregate variables helps improve macro forecasts of real economic activity and find that aggregate changes in profit margins as well as in asset turnovers are leading indicators of real GDP growth. Bonsall et al. (2013) document that firm-specific management forecasts are significantly related with aggregate market returns around the forecast release dates and conclude that management forecasts provide timely information to the market about the macroeconomy. However, none of these studies focuses on explaining the predictive ability of aggregate earnings for future inflation shocks, which is the primary focus of this study. 2.2 Hypotheses development The next two sub-sections present alternative mechanisms through which aggregate earnings potentially affects future inflation shocks with an aim to establishing causality between aggregate earnings and future inflation news one based on the demand for investment products and the other based on demand for consumption goods Investment Demand hypothesis 8

10 The investment demand hypothesis proposes that aggregate earnings shocks cause an increase in firms demand for investment goods and services and that this increase in the short-run, when supply of production goods is relatively inelastic, leads to an increase in prices of production goods. Changes in investment decisions from unexpected profitability can affect corporate demand for raw materials, inventory and capital goods/services, which in turn raises the prices of these inputs, at least in the short-term, where the supply of these goods and services is less than perfectly elastic. Unexpected changes in corporate profitability are likely to affect future investment in an economy for at least three reasons. First, current profitability, through the persistence of earnings, is indicative of the profitability of future investments. Corporate profitability thus directly affects the perceived attractiveness of investment opportunities, which in turn influences managers investment decisions. Second, greater profits facilitate greater investments by loosening financing frictions such as the cost of raising equity capital and the costs of debt overhang. Specifically, to the extent that profits are quickly converted into cash, higher profits translate into increased availability of internal funds (see Hennessy, Levy and Whited, 2007; Lewellen and Lewellen, 2012) and lower reliance on external funds for investments. Along similar lines, lower profits increase a firm s reliance on external funds for investments. Third, corporate profitability and bank lending are also positively inter-related. Improved profitability lowers perceived credit risk of a business (e.g., Altman, 1968), which contributes to increased bank lending. The increased availability of credit that flows into corporate investments leads to greater demand for production goods and services, which in turn, increases inflation. Consistent with this effect of bank lending on investments, Bassett et al. (2010) document a significant increase in real GDP and inflation in the quarters immediately following a shock that relaxes bank lending standards. Greater profitability can lead firms to invest both in current operations by increasing inventory levels or alternatively investing in operating capacity for the future through increased capital expenditures. Similarly, firms may respond to unexpected declines in profitability by cutting 9

11 their inventory or delaying or even cancelling proposed capital expenditures. In our discussions, we use the term production goods to refer to both goods held for resale by a firm as well as goods and services employed by a firm to build operating capacity. Thus, the main prediction of the Investment Demand Hypothesis can be formally stated as: H1: Unexpectedly high (low) corporate profits cause demand for investments to increase (decrease) and in the near term cause prices of production goods to increase (decrease) Consumption Demand hypothesis Consumption demand hypothesis states that shocks to aggregate profitability changes consumption demand by affecting individual s wealth and this shock to demand, in the short-run when supply of consumption goods is not perfectly elastic, causes the prices of consumption goods to change. At firm-level, greater profits generated by a firm increase returns to firm s investors either through higher dividends or through heightened share and bond prices. Also, since firms are known to share profits with employees through performance-related bonuses and commissions, shocks to firm s profits directly affect the disposable income of employees. For instance, Blanchflower et al. (1996) document that movements in profitability are followed by changes in workers remuneration and that the elasticity of wages with respect to profit-per-employee is Along similar lines, Bronars and Famulari (2001) document that a 4% increase in market value of a firm s equity raises employees pay within the next three years by 0.3%. The above arguments imply that unexpected increases in firm s profitability cause an increase in net wealth of individual agents in an economy, which if at least partly consumed in the same period leads to an increase in short-run prices for consumption goods and services, due to relatively inelastic supply in the short-run. Supporting the view that agents tend to consume part of unexpected increases in wealth, Davis and Palumbo (2001) estimate that each dollar of unanticipated stock price increase causes consumer spending in the contemporaneous period to increase between 4 and 7 cents. This argument links aggregate earnings to future inflation shocks through increased consumption by individuals and leads to the following prediction: 10

12 H2: Unexpectedly high (low) corporate profits cause demand for consumption to increase (decrease) and in the near term cause prices of consumption goods to increase (decrease). Overall, if the primary effect of aggregate earnings on future inflation occurs through its effect on aggregate demand for production goods, we expect corporate earnings to be positively related to future shocks in demand for and prices of production goods, while we expect a positive relation between aggregate earnings and future demand for and prices of consumption goods if the primary effect of aggregate earnings on future inflation occurs through consumer spending. 3. Inflation Measures Tests of H1 and H2 require measures for price of production goods and of consumption goods. Towards this, we use the Producers Price Index (PPI) to track inflation in prices of production goods, as the main objective of PPI is to track prices of goods and services from a producer s perspective. As discussed below, the goods and services tracked by PPI are the ones that are primarily purchased by other businesses. We use the Consumer Price Index (CPI) to capture prices of consumption goods, as the main objective of CPI is to track prices paid by final consumers. 9 The prices of production goods (i.e., Producer Price Index) are measured at three different points in the production chain, giving rise to three related measures of PPI: (i) PPI for crude materials, (ii) PPI for intermediate goods and (iii) PPI for finished goods. Crude materials are unprocessed materials that are inputs for production of intermediate goods. Intermediate goods are either partially processed goods or fully processed goods that are bought by businesses for input in their operations. Finally, finished goods are commodities that are ready for sale to other businesses or consumers without any further processing. In contrast to these indices, the Consumer Price Index captures prices at the point of purchase by end consumers. Thus, for example, the Bureau of Labor Statistics (BLS) 9 Although in the long-run, price shocks to production goods could affect prices of consumption of goods, our focus is only on short-term price effects to unexpected changes in corporate profitability. Studying long-term effects of aggregate earnings on macroeconomy requires a comprehensive description of the macroeconomy, which is a tall order. As Fisher (1953, p 21) observes to construct this sort of a (comprehensive) model in some theoretical sense is difficult enough; but then to attempt to treat it statistically brings up further obstacles, especially in view of the limitations imposed by the analytical tools and basic data Analysis of short-term effects helps partly get over these constraints by allowing researchers to focus on shocks to the variables of interest. 11

13 includes price of iron-ore in PPI for crude materials, price of steel mill products that are produced with the iron-ore in PPI for intermediate goods, price of metal-forming machine tools that rely on steel products as inputs in PPI for finished goods and finally the price of new vehicles that use metalforming machine tools as capital equipment in CPI. We discuss below in greater detail differences in the construction of CPI and PPI. 3.1 CPI versus PPI 10 The CPI is the most widely used measure of inflation and is often viewed as the benchmark inflation guide for the U.S. economy. The CPI focuses on tracking price changes affecting consumers and primarily measures adjustments needed to consumer payments when the intention is to maintain consumers purchasing power. The CPI is derived from prices on a base-year market basket that is selected by weighting 200 categories of goods and services grouped under eight categories: food and beverages, housing, apparel, transportation, medical care, recreation, education and communication and other goods and services. The weight for each category in the CPI-basket is based on expenditure information gathered through surveys of thousands of families across the country. 11 For each of the categories in the basket, the Bureau chooses samples of several hundred specific items within selected business establishments and doctors offices. The prices considered for the index include taxes and government fees directly associated with the purchase of goods and services, such as sales and excise taxes. In contrast to the consumer-oriented focus of CPI, the Producer Price Index (PPI) measures the average change over time in the selling prices received by domestic U.S. producers for their output. 12 The prices included in the PPI are from the first commercial transaction for products and 10 The discussions in this sub-section are largely based on information provided in the Bureau of Labor Statistics website. 11 The expenditure weights are currently updated every 2 years. Prior to 2002, the weights were updated roughly once every 10 years. 12 Aggregate earnings of U.S. listed firms include earnings from products sold outside the U.S. The effect of such foreign earnings on the PPI depends on whether the product is manufactured in the U.S. or not. Our discussions with officials at Bureau of Labor Statistics reveal that, if a product is manufactured in the U.S., then the prices of such products are considered for PPI, but not otherwise. In our analysis, we considered excluding foreign earnings from the definition of aggregate earnings. Given that some of the foreign earnings are relevant for PPI, a priori it is not obvious that foreign earnings should be excluded from aggregate earnings. In any case, we could not conduct our analysis on domestic earnings, as segment data on domestic earnings is available only 12

14 services and reflect the net revenues received by producers. These prices, therefore, exclude revenues collected on behalf of governments, such as sales taxes, and the effects of customer-rebates or subsidized-interest financing if the costs of these promotions are not borne by the manufacturers. 13 The PPI index is broader in its coverage of goods and services than CPI and primarily covers goods and services purchased by other businesses, which justifies our use of PPI to measure the price changes for production goods. But since PPI also includes goods and services bought by consumers from retail sellers and directly from producers, we later check sensitivity of results linking aggregate earnings news to PPI by separately analyzing the link between earnings news of non-retail firms with PPI. Also, unlike CPI, PPI includes price changes for goods produced in the U.S. specifically for military or export, but does not track changes in prices of imported goods and services. 3.2 Alternative inflation measures Apart from CPI and PPI, there are other measures of inflation such as GDP deflator (which tracks the combined price changes facing governments - federal, state and local, businesses and consumers), employment cost index (which measures changes in labor costs in private and governmental sectors), employer costs for employee compensation (which tracks the average cost to employers for wages and salaries and benefits per employee hour worked). While these alternative measures of inflation could track effects of aggregate earnings news on prices of goods and services for the overall economy and to labor costs, these inflation measures are released only at a quarterly frequency, which, compared to monthly macro series, is likely to lower the ability to identify immediate price effects of earnings shocks. This is of particular concern as our analyses control for a variety of macro and financial variables and use of quarterly inflation measures allows more time for control variables to soak-up the inflation-information in aggregate earnings news. Consistent with this concern, we find little predictive relation between aggregate earnings news and future inflation variables measured at a quarterly frequency in unreported results. on an annual basis in Compustat, whereas our analysis requires data at a monthly or at most quarterly frequency. Also, analysts forecasts for domestic earnings are unavailable to compute surprises on domestic earnings. 13 The PPI views wholesalers and retailers as service providers (such as marketing and distributing goods to customers) rather than suppliers of goods. Hence, the PPI computes output-prices for wholesale or retail trade establishment as the difference between selling price of a good and its acquisition price. 13

15 Although BLS reports several components to each headline inflation index, we focus on the headline indices, as these tend to be the broadest and most comprehensive measures and most closely correspond to measurement of aggregate earnings across all sectors in the extant literature. This is also in line with prior studies that have typically focused on headline CPI to proxy for inflation. Thus, our measure of CPI is the Consumer Price Index for All Urban Consumers (CPI-U). Similarly, for PPI, it is the PPI Finished Goods, PPI Intermediate Goods and PPI Crude Materials. 4. Sample Selection and Data Description 4.1 Sample selection and variable definitions Our initial sample consists of all NYSE, AMEX and NASDAQ firms in the merged CRSP/Compustat database, with data available on earnings announcement dates. From this sample, we include only the firms available on IBES. For each quarterly earnings announcement, we use the most recent IBES consensus analyst forecast prior to the announcement to calculate firm-level earnings surprises. Specifically, a firm-level earnings surprise is calculated as IBES-reported quarterly earnings minus the most recent consensus analyst earnings forecast from IBES, divided by the absolute value of the most recent consensus analyst earnings forecast. We calculate aggregate earnings surprises by taking value-weighted and equal-weighted averages of firm-level earnings surprises over a quarter (three months). In particular, VIBES (EIBES) is the quarterly value-weighted (equal-weighted) average of IBES earnings surprises. In calculating aggregate earnings surprises, we exclude all firms with share price less than $1, and earnings surprises in the top and bottom 0.5% of the empirical distribution. We obtain quarterly data on corporate investment and personal consumption from Federal Reserve Economic Data (FRED) system of St. Louis Fed over the period from January 1985 to June The starting date of this sample period is governed by the availability of IBES data to construct aggregate earnings surprises. To proxy for investments at the macro-level, we employ the following four alternative measures: (i) Quarterly percentage change in gross private domestic investment (AGPDI), which measures physical investment in the productive capacity of the economy. It includes replacement purchases, addition to capital assets net of depreciation and investment in inventory. (ii) 14

16 Quarterly percentage change in private non-residential fixed investment (APNFI), which captures domestic spending by private sector enterprises and households on commercial or industrial buildings and other commercial or industrial structures with long economic lives. It specifically excludes investments in residential homes and other residential structures. (iii) Quarterly percentage change in gross fixed capital formation (GCFC), which reflects improvements to land, purchases of plant, machinery, and equipment and construction of infrastructure, such as roads and railways, commercial and industrial buildings and private residential dwellings. (iv) Quarterly percentage change in aggregate inventories (ACINV), which is computed by aggregating inventory of listed firms on Compustat and then computing the quarterly changes in aggregate inventory. 14 To measure consumption at the macro-level, we focus directly on consumption outlays and expenditures as well as indirectly on consumption by analyzing changes in personal income. To the extent that consumers do not entirely save changes in their personal income, changes in income would reflect changes in consumption. 15 Our proxies for consumption expenditures and personal income are: (i) Quarterly percentage change in personal outlays (APO), which primarily tracks personal consumption on goods and services, interest payments by households on non-mortgage debt and transfer payments to government or social services. (ii) Quarterly percentage change in personal consumption expenditures (APCE), which is a component of personal outlays and consists of household expenditures on durable and non-durable goods and services. (iii) Quarterly percentage change in personal income (APINC), which measures income received by individuals from all sources, including wages and salaries, dividends and interest, rental income, and the like. Personal Income also includes income received by non-profit institutions serving households, by private noninsured welfare funds, and by private trust funds, but excludes gains from sale of assets and pension benefit payments. (iv) Quarterly percentage change in disposable personal income (ADPINC), which is defined as Personal Income less personal tax payments. 14 Our conclusions remain unaffected when we use the change in private inventory reported by the Bureau of Economic Analysis instead of aggregate inventory from Compustat. 15 Hyndman and Athanasopoulos (2013) estimate that a 1% change in personal disposable income results in an average increase of 0.84% in personal consumption expenditure. 15

17 Although data on personal consumption and personal income are available at a monthly frequency, we report results based on the quarterly data to be consistent with the VAR results for investment proxies. However, our conclusions remain unchanged, if we conduct these analyses using monthly observations. To allay concerns that any observed relation between aggregate earnings surprise and future investment, consumption or inflation is confounded by omission of correlated macroeconomic information, our analyses additionally include controls for three interest-rate-spread variables, a macro-variable to capture real output and equity market return. The financial variables are included as they reflect forward-looking information about the macroeconomy. Specifically, the following five control variables are included in our analyses: (i) Change in percentage default spreads, calculated as the difference in interest rates between AAA bonds and BAA bonds in period t ( DEFAULT t ), (ii) Change in percentage yield spreads, calculated as the difference in interest rates between federal funds rate and risk free rate, in period t ( YIELD t ) (iii) Change in percentage term spreads, calculated as the difference in interest rates between 10-year government bonds and risk free rate, in period t ( TERM t ), (iv) Percentage change in industrial production index released in period t (PROD t ) and (v) Percentage change in S&P 500 index in period t (SP500 t ). 16 Throughout, macro-variables are measured in the period in which they are released to be consistent with the timing of the measurement of the aggregate-earnings variable. We obtain inflation data from Bureau of Labor Statistics. PPI and CPI numbers are generally released by BLS at 8:30 a.m. EST, before stock markets open. For most months in the sample period, the PPI announcements precede the CPI announcements. Since inflation data are available at a monthly frequency and since price-effects are likely to be short-lived when demand or supply of goods and services is relatively inelastic, we conduct analyses of inflation variables at the monthly frequency. In these analyses, to ensure that the aggregate earnings news are representative of all listed firms, for each month t, we computed value-weighted and equally-weighted aggregate earnings news (VIBES t-1,t-3 or EIBES t-1,t-3 ) using all earnings announced over a rolling 3-month window from t-1 to t- 16 S&P 500 index data are from Center for Research in Security Prices (CRSP). 16

18 3. To be consistent, our macro and financial control variables are also measured over the same 3- month window, although our results are robust to measuring these variables over the last month alone (i.e., over month t-1) Summary statistics Panels A and B of Table 1 reports summary statistics for the variables used in the quarterly and monthly analyses respectively. When we look at quarterly statistics, the mean aggregate earnings surprise based on analysts earnings forecasts is 5.8% for the value-weighted measure (VIBES) and 13.0% for the equal-weighted measure (EIBES) during the sample period,. These averages are significantly different from zero, as are the median aggregate earnings surprises, reflecting the wellestablished optimistic bias of analysts (e.g., Brown, Foster and Noreen, 1985). 18 Our sample period is characterized by growth in investment, consumption and personal income in the U.S. Aggregate inventory (ACINV) exhibits the largest average quarterly growth of 1.6% while gross fixed capital formation (GFCF) has the lowest average quarterly growth of 0.7%. The averages as well as medians are significantly greater than zero. Among the quarterly control variables, the mean and median values for DEFAULT q, YIELD q, and TERM q are insignificantly different from zero, suggesting that there was no trends in these interest-rate variables during the sample period. The average production index increased by 0.54%, and average S&P 500 index increased by 2.4% during our sample period. Not surprisingly, summary statistics for monthly aggregate earnings as well as monthly control variables in Panel B are similar to those reported in Panel A. The average CPI and PPI inflation rates during the sample period were 0.23% and 0.19% per month. Table 2 reports Pearson correlations (above the diagonal) and Spearman correlations (below the diagonal) among the variables of interest. Panel A reports the correlations for variables employed in the quarterly analyses and Panel B for those in the monthly analyses. Not surprisingly, the equally- 17 In month analyses, since the variables are measured over more than one month, we include two subscripts to the variables to indicate the starting month and ending month over which they are computed. 18 The extant literature on analysts forecasts documents a variety of biases and inefficiencies in analysts earnings forecasts. To the extent that biases in analysts forecasts induce noise in our earnings surprise measures for the aggregate market, our analyses would be conservatively biased. Nonetheless, in subsequent sections, we check the robustness of our results to time-series based measures of earnings surprises. 17

19 weighted and value-weighted measures of aggregate earnings are highly correlated. These aggregate earnings surprises are also significantly positively correlated with changes in yield spreads as well as with industrial production index. Also, the actual PPI and CPI measures of inflations are highly correlated with each other. The correlation coefficients across these alternative measures of inflation exceed 0.6, suggesting a large overlap in the price increases measured using PPI and CPI. 5. Empirical Results 5.1 Vector auto-regression analysis As is common in the macroeconomic literature, we study the macroeconomic information content of aggregate earnings surprises using vector auto-regressions (VAR). This approach addresses potential endogeneity in inter-related macroeconomic variables and allows us to investigate the information content of aggregate earnings surprises for future innovations to corporate investment and personal consumption. Specifically, we estimate the following VAR model: = + + (1) where z q is a vector. The variables included in the vector are: (i) percentage change in corporate investment proxy released in quarter q (AGPDI, APNFI, GFCF or ACINV), (ii) percentage change in personal consumption or income proxy released in quarter q (APO, APCE, APINC or ADPINC) (iii) aggregate earnings surprise released in quarter q (VIBES or EIBES), (iv) change in percentage default spreads (the difference in interest rates between AAA bonds and BAA bonds) in quarter q ( DEFAULT q ), (v) change in percentage yield spreads (the difference in interest rates between federal funds rate and risk free rate) in quarter q ( YIELD q ), (vi) change in percentage term spreads (the difference in interest rates between 10-year government bonds and risk free rate) in quarter q ( TERM q ), (vii) percentage quarterly change in industrial production index (PROD q ) released in quarter q, and (viii) percentage quarterly change in S&P 500 index (SP500 q ) released in quarter q. All the variables in the VAR system appear to be stationary based on the Phillips-Perron tests for unit root. Our use of 2 lags in the system is identified based on the Final Prediction Error Criterion. Using the standard approach for presentation of VAR results, Table 3 presents impulse response functions (IRFs) from estimating Equation (1) along with 95% confidence intervals and 18

20 standard errors. The IRFs describe how a response variable reacts to a unit (standard-deviation) positive shock to an impulse variable over time. Although the VAR model includes a variety of variables and many lags of each variable, to conserve space, we report only the impulse responses of the variable of interest (proxies for investment, consumption or income) to aggregate earnings shocks for up to four-periods ahead. From Panel A, we find that aggregate earnings surprises predict innovations in gross private domestic investment up to one-quarter-ahead. Consistent results are also obtained for other proxies for investments, viz., changes in private non-residential fixed investments, changes in gross fixed capital formation or changes in aggregate inventories. For all investment proxies, we find that shocks to profitability affect investments in the subsequent one to three quarters. However, in Panel B, when we focus on the results from VAR analysis of consumption or personal income, we do not find any evidence that aggregate earnings surprises affect growth in future personal outlays, personal consumption expenditure, personal income or personal disposable income. Collectively, the VAR results show that aggregate earnings surprises significantly predict only future corporate investment innovations, but not future consumption innovations. The predictive ability for investment innovations is particularly interesting as our VAR analysis accounts for both potential persistence in actual corporate investment and controls for effects of a broad set of financial and macroeconomic control variables, implying that the predictive ability of aggregate earnings is incremental to that of the predictive content of these control variables. Also, the failure to find predictive power of aggregate earnings surprises for personal consumption innovations, suggests that unexpected corporate profits do not affect consumption within at least the subsequent four quarters. These findings support the investment demand hypothesis, but not the consumption demand hypothesis. 5.2 Firm-level investment analysis The VAR analysis provides strong evidence that aggregate earnings shocks impact innovations to future corporate investment at the macro level. To get a deeper understanding of the source behind this macro-level link, we study whether individual firms change their investment behavior in response to profitability shocks. Evidence of a relationship at the firm-level provides 19