Looking Inside the Magic 8 Ball: An Analysis of Sales Forecasts using Italian Firm-Level Data

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Looking Inside the Magic 8 Ball: An Analysis of Sales Forecasts using Italian Firm-Level Data Maria D. Tito 2017-027 Please cite this paper as: Tito, Maria D. (2017). “Looking Inside the Magic 8 Ball: An Analysis of Sales Forecasts using Italian Firm-Level Data,” Finance and Economics Discussion Series 2017-027. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2017.027. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Looking Inside the Magic 8 Ball: An Analysis of Sales Forecasts using Italian Firm-Level Data Maria D. Tito∗ February 28, 2017 Abstract Thispaperexploresfirmforecastingstrategies. UsingItaliandata,wefocusontwoaspects of the forecasting process: how firms forecast sales and how accurate their predictions are. We relate both outcomes to current conditions, firm experience, global factors, and other firm characteristics. We find that current conditions tend to explain most of the variability in the sales forecast. While past projection errors tend to account for cross-firm differences in models of expectation formation, they are a key explanatory variable in models of forecast accuracy. Among other controls, firm size, experience, and global conditions–through the effect of price changes that the firm anticipates–shape firm expectations and influence the projection errors. Our findings suggest that models of sales expectations should take firm characteristics and market heterogeneity into account. Key words: Exporting, Sales Forecasting, Forecast Accuracy. JEL classification: F14 1 Introduction Sales forecasts are a crucial tool for business planning, marketing, and general decisionmaking, as more accurate forecasts improve firm performance. Understanding how firms predict sales is also a premise for empirical research, where estimating the parameters associated with various firm’s ∗BoardofGovernorsoftheFederalReserveSystem,20thStreetandConstitutionAveNW,Washington,DC 20551. Contact: maria.d.tito@frb.gov. WewouldliketothankFelixTintelnotfortheusefulinsightsandcomments. ResultsforthisprojectwereobtainedthroughtheBankofItalyRemoteaccesstomicroData(BIRD).Theviews presentedinthispaperrepresentthoseoftheauthoranddonotnecessarilycoincidewiththoseoftheFederalReserveSystemortheBankofItaly. 1

problems depends on the firm sales forecast. A firm’s decision to export, for example, depends on its profit expectation in foreign markets; similarly, developing new products requires projecting future demand. While many contributions assume that firm’s sales expectations depend on some setofobservables,theexactfunctionaldependenceinfluencestheestimateofstructuralparameters. In particular, Dickstein and Morales [2015] show that the sensitivity of export flows to transport costsisheavilydependentonhowtheresearcherspecifiesthefirm’sforeignprofitexpectation.1 The authors propose a strategy to recover bounds for the fixed costs of exporting. This methodology has the advantage of requiring only partial knowledge on the conditioning set of information; the bounds, however, remain conditional on a subset of variables that firms employ to predict export revenues–specifically,theauthorsassumethateachfirmknowsonlytheaggregateexportsfromChile to each destination market in the previous year, the distance to each market, and the firm’s own productivity in the previous year.2 Further progress on firm information sets may be achieved by directly analyzing data on expectations. We draw on a novel dataset, the Survey of Industrial and Service Firms (INVIND), which collects data on expectations for domestic and foreign sales, to explore firm forecasting strategies. In particular, we focus on two aspects of the forecasting process: how firms forecast sales and how accuratetheirpredictionsare. Ourmodelsrelatefirmexpectationsandforecasterrorstofourindicators: currentconditions,firm’sexperience,globalfactors,andotherfirmcharacteristics. Weanalyze sales predictions by destination market, distinguishing between domestic and foreign markets. We find that current conditions tend to explain most of the variability in the domestic and foreign sales forecast;inparticular,currentsalesexplainaround75percentofthewithin-firmforecastvariability. If we take also the cross-sectional variation into account, the R2 of a regression that includes only current sales increases to more than 95 percent. While past projection errors tend to account for cross-firm differences in models of expectation formation, they are a key explanatory variable in models of forecast accuracy. Among other controls, firm size, experience, and global conditions–through the effect of price changes that the firm anticipates–shape firm expectations and influence the projection errors, with 1Thisproblemiscommontoallsettingsinwhichagents’decisionsdependontheirexpectations. Manski[1993], Manski[2004],andCunhaandHeckman[2007]highlighttheinfluenceofspecifyingagents’expectationwhenevaluatingreturnstoschooling. 2DicksteinandMorales[2015]proposeatesttoanalyzethecontentofpotentialexporters’informationsets. Theprocedure,however,canbecomecumbersomeaseachadditionalvariableincludedinthesetwouldrequirea newimplementationofthetest. 2

magnitudes varying across markets. Our paper belongs to the literature on sales forecast, recently reviewed by Winklhofer et al. [1996]. Earlier contributions focused on the forecast accuracy and collected cross-sectional survey data on firm-level projection errors. Our paper relies on a novel data source that directly records data on expectations and offers cross-sectional and panel variation. A similar data source exists for Japanesefirms, butithasnever, tothebestofourknowledge, beenusedtoanalyzefirmforecasting strategies.3 Our paper provides an empirical microfoundation to sales expectations models. While the literature has mostly adopted either perfect foresight or limited information models of expectation formation, with the latter assuming that firms use only specific observables to predict sales, our results point to the empirical importance of current sales, firm characteristics, and prices as factors affecting the forecast.4 Moreover, our models stress the interplay between market characteristics and expectations. The rest of the paper is organized as follows. Section 2 describes the data and presents some preliminary evidence. Section 3 develops an empirical model of sales forecast. Section 4 identifies the variables that influence the accuracy of the forecast. Section 5 concludes. 2 Data and Descriptive Analysis The empirical analysis draws upon the Survey of Industrial and Service Firms (INVIND), a representative panel of Italian firms with at least 20 employees.5 The data are collected by the Bank of Italy,throughitsregionalbranches;thiscollectionmethodensuresahighresponserate–77percent, on average – and a rigorous control on reported answers. The survey collects data on standard balance sheet items (wage bill, employment, sales, investments, etc.) and on one-year-ahead expectations of domestic and foreign sales. In each year, firms 3TheBasicSurveyofOverseasBusinessActivitycollectsdataonsalesandinvestmentexpectationsfor Japaneseparentsandtheirforeignaffiliates;see,forexample,Chenetal.[2016]. Qualitativeinformationonfirm currentbusinessconditionsandexpectationsareavailablethroughtheIFOBusinessClimateSurveyforGermany andthePhillyFed’sBusinessOutlookSurveyfortheUnitedStates;see,forexample,Bachmannetal.[2013]. 4Ourfindingsareconsistentwithotherpapersinthetradeliterature. Theimportanceofexperienceis,forexample,stressedbyAlbornozetal.[2012],whodevelopamodelofexperimentationinwhichexporterslearnfrom previousexperience. Moreover,DicksteinandMorales[2015]cannotrejectthehypothesisthatexportersknowat leastdistance,theirownlaggeddomesticsales,andlaggedaggregateexportswhenmakingtheirexportdecisions. 5Bancad’Italia,IndaginesulleImpreseIndustrialiedeiServizi,1995-2012. Until2000,thesurveywasrestricted tofirmswithaminimumof50employees. AbriefdescriptionisincludedinTito[2015]. 3

are asked to report their sales expectations for domestic and foreign markets. Export data are available as aggregates across all foreign markets in which a firm operates; the only exception is the 2008 survey, which records a breakdown of firm foreign sales and expected shipments by region (European Union, North America, China, and Rest of the World).6 Our analysis focuses on manufacturing firms with at least 50 employees; moreover, as data on expectations are available only since 1995, our sample covers 1995-2015 and includes 24,500 observations (here an observation is a firm-year combination) from the initial 67,417 records. Figures1and2offerapreliminaryviewonthedataonexpectations,comparingsalesprojections to current shipments. Expected sales mirror the path of current sales, hinting to a random-walk behavior: Firms, on average, expect their future revenues to be roughly in line with the realized sales. The wedge between current and expected sales is larger on the domestic market (figure 1) where factors influencing demand other than those embedded in current shipments are likely easier to identify. Conversely, the hurdle of acquiring knowledge on market conditions abroad may induce firms to closely shape the path of expected sales on that of current shipments (figure 2). Sales expectations inherit the properties of current sales, with foreign shipments representing around 40 percent of total shipments and displaying larger variability, as shown in table A1.7 Tables 1 and 2 highlight cross-sectional differences. We construct the growth rates of domestic and foreign sales and compare current rates to the forecast; the results are shown for 2005. While firm expectations tend to be, on average, in line with future growth, the projections appear more concentrated in foreign markets. Firm assign a negligible probability to export growth less than -10 percent or higher than +20 percent; the interval is wider for the domestic market. As for the time-series evidence, this difference may reflect the larger information costs firms face abroad. Finally, figure 3 and 4 look at the forecast in revision space. Firms tend to consistently overpredict their sales; the forecast error, the difference between realized and expected sales in any given year, implies that firms do not fully know their future revenues when making a decision on their operations. Excluding 2001, when the introduction of the euro currency promoted conditions favorabletoentryinEuropeanmarkets,therevisionsare,onaverage,largerforthedomesticmarket forecast. This result is not surprising for two reasons. First, trade policy and foreign regulations likely contributed to tapering down sales fluctuations in foreign markets, improving the accuracy of 6Dataoncurrentshipments(butnotonexpectations)byregionarealsoavailableinthe2006survey. 7Whiletheshareofforeignsalesoutoftotalshipmentsisontheupperboundofthevaluesreportedbythe tradeliterature,ourdatasetdoesnotincludefirmswithfewerthan50employees. 4

firm export forecasts. Second, selling to different markets could work as a diversification strategy: projection errors would compensate if fluctuations in demand were negatively correlated across markets. 3 Forecasting Sales: Domestic vs. Foreign Markets How do firms forecast sales? Figures 1-2 flag current sales as an important predictor to firm forecasts. The discrepancy between expected and current values, however, suggests that other factors may influence the projection. We propose a baseline model that relates expected sales to current shipments and past forecast errors, lnE [Sales ]=β +β ·lnSales +β ·Forecast Error +D +ε (1) t i,t+1 0 1 it 2 it st it wherelnE [Sales ] denotesthet+1valueofsalesexpectedbyfirmiattimet,lnSales represent t t+1 it it current shipments, and Forecast Error captures the projection error, defined as the difference it between current sales and past projections (in log-s).8 Our baseline model focuses on β and β . β 1 2 1 identifiesallfactorsthatinfluencetheprojectionthroughchangesinsales; aslnSales isasufficient it statistics for firm productivity, β captures also the effect of productivity shocks on the forecast. 1 The forecast error introduces a mean-reverting component to the process of expectation formation; we expect β > 0 as firms would boost their projections if realized sales were above expectations 2 and viceversa. Weestimatemodel(1)separatelybymarket;althoughourapproachdoesnotallowustodirectly comparethemagnitudeofthemarginaleffectsandtheirimplicationsontheprojection, ourinterest mainly lies in highlighting the qualitative differences between the domestic and foreign market forecasts. In our sample, we set (log) foreign shipments to zero for firms operating only on the domestic market, while we exclude firms that report zero sales in the home market. More precisely, (cid:0) (cid:2) (cid:3) (cid:1) we define expected foreign sales as ln 1+E Exports and foreign sales as ln(1+Exports ) t t+1 it it to exploit the variation in sales due to entry in a foreign market. Instead, we define expected 8Inourmodel,wespecifytheforecasterrorasasimpledifference,asboththesizeandthesignoftheforecast errormatterfortheexpectationformationinourempiricalfindings. Inunreportedresults,theabsolutedeviation oftherealizedsalesfromtheexpectationsdoesnotsignificantlyaffecttheprojection. 5

domestic sales as lnE [Dom. Sales ] and domestic sales as lnDom. Sales .9 t t+1 it it In addition to current sales and projection errors, our decomposition explores the role of four potentialpredictors: experience, currentconditionsinothermarkets, pastsales, andfirmcharacteristics. Following the literature on forecasting, we include the share of sales in the market and firm agetocapturelearningaboutmarketconditions.10 Ourmodeltestswhetherfirmsextractsignalson futuresalesfromconditionsinothermarketsorfrompastsalesbyadding,forexample,pricechanges experienced by the firm across all markets, foreign shipments, and past sales to our list of variables in the home market regression. We include total employment to control for factors varying at the firmlevelandaffectingtheforecast–forexample, cross-firmdifferencesintheforecastbudget.11 In some specifications, we add firm fixed effects to absorb all time-invariant firm characteristics that influence the projection. Finally, we absorb sector-specific revenue shifters by using sector-time dummies. Results Thissectionanalyzestheestimatesofmodel(1)bymarket. Tables3and4reporttheOLSregression results; table 3 presents the estimates for the domestic market, while table 4 focuses on foreign markets. Theresultsrevealkeysimilaritiesacrossmarkets. Currentsalesarethemostimportantpredictor ofdomesticandforeignsalesexpectations;ineachcase,theyaccountformorethan95percentofthe overallforecastvariability. ThecoefficientonlnSales suggeststhatlargersalesinthecurrentperiod it induce the firm to improve the near-term outlook at home and abroad. Looking at magnitudes, a one-standard-deviation increase in sales, which correspond to an increase in domestic sales by 5,000 euroandinforeignsalesby55,000euro,isassociatedwithaone-standard-deviationupwardrevision to the domestic and foreign market forecast.12 Inbothmodels,theeffectofsalesextendsbeyondthatofcurrentshipments: ourestimatesimply that firms tend to soften their expectations if current results exceed past projections. Similarly, we find a negative correlation between sales at t−1 and expectations for t+1. 9Byconstruction,allfirmsinoursampleoperateonthedomesticmarketbutnotnecessarilyontheforeign market. 10See,forexample,Cavusgil[1984]andDiamantopoulosandWinklhofer[1999]. 11Recentstudiesconcludethattheaccuracyofforecastisinfluencedbyfirmsize. See,forexample,Dalrymple [1987]. 12SeeTableA1forthesummarystatisticsofthemainvariablesofinterest. 6

The impact of other components varies by market. The domestic market forecast is significantly influenced by experience and firm characteristics. Among proxies of experience, the negative and significant effect of Age indicates that older firms tend to be more conservative in their projections; less robust implications, instead, are associated with the coefficient of Dom. Share, which becomes positive in the last two columns, after introducing indicators of past outcomes. Firm characteristics are positively correlated with expectations, implying that bigger firms are more optimistic about the projection. Moving to foreign market estimates (table 4), we find that firms take signal from errors to the domestic forecast: as with errors on foreign projections, revisions to the domestic forecasts are negatively associated with expectations of foreign sales. The significant effect of domestic sales, through its deviation from expectations, is consistent with the findings in Dickstein and Morales [2015], which fail to reject the inclusion of past domestic sales in the information set of potential exporters when deciding whether to start selling their products abroad. Tables5and6isolatethecontributionofthewithin-firmvariation. Theintroductionoffirmfixed effectsreducesthesimilaritiesbetweenthedomesticandforeignmarketspecifications. Currentsales remain the main predictor of the firm forecast; however, it explains a smaller share of the variation in projections if compared with the results in tables 3 and 4–namely, around 75 percent in FE regressions vs. more than 95 percent in the OLS regressions. Reassessing magnitudes, increasing currentsalesbyonestandarddeviation(sd)raisesthedomesticmarketprojectionby75percentofa sd and the foreign market projection by 80 percent of a sd. Thus, removing idiosyncratic cross-firm differences weakens the effect of current sales on the forecast. Other similarities are limited to the role played by past sales, firm size, and experience. In tables5and6, differentlyfromtheOLSresults, pastsalesarepositivelyandsignificantlycorrelated with the projections. Employment shows a similar positive correlation, with bigger firms predicting significantly larger gains in future sales at home and abroad. While proxies of experience share similar signs in tables 5 and 6, they significantly influence only the domestic sales forecast through the effect of Dom. Share, which preserves a positive coefficient. Controlling for cross-firm differences alters the effect of forecast errors from the OLS results. While projection errors tend to lose significance in the domestic market specification when adding all controls, they are positively correlated with foreign market expectations in the last two columns 7

of table 6; this result, however, is not robust across specifications. Global conditions offer other points of contrast across markets. First, while domestic conditions significantly contribute to modulating expectations of foreign shipments, outcomes in the export market do not significantly affect the domestic outlook; this difference may be related to the higher cost of acquiring information on foreign consumers. Second, higher prices are viewed as a signal of larger revenues at home, but of lower revenues abroad; this result supports empirical findings of higher elasticity of demand for foreign goods. 3.1 Forecasting Export Sales by Market The heterogeneity across foreign markets complicates the forecast of export sales and the interpretation of our results.13 Exploiting the 2008 survey, which separately records current and expected shipments to the European Union, North America, and China, we estimate models of sales forecast for each region. To control for the endogenous selection into a particular market, we restrict our sample to the set of firms that export to all three regions: This constraint substantially reduces the number of firms in our sample to 885. Results are shown in tables 7-9. In all three markets, current sales remain the best predictor of futureshipments. ForNorthAmerica,however,evenaftercontrollingforcountry-specificsector-year revenue shifters, other factors also exercise a significant influence on the forecast. Global conditions achieve diverging effects on the near-term sales outlook. On the one hand, overall projection errors raise future sales expectations. On the other hand, changes in prices are negatively related to the sales forecast: Higher prices abroad translate into lower revenues, due to the higher elasticity of demand relative to the domestic market. Cross-firm differences in employment also affect the projection, with bigger firms lowering their North American sales forecast. Our identification is based on cross-firm variation within sector-year cells. The exclusive availability of 2008 data threatens the general validity of the model because of special factors related to the Great Recession. Including controls for the severity of the crisis does not change our results (final column in tables 7-9); however, we cannot exclude the influence of other factors. 13SeeAnderson[1960]foradiscussionoftheproblemsspecifictotheexportsalesforecast. 8

4 Revisions to Forecast We examine the sales forecast error using a model that relates the difference between expected and realized values to past errors and past sales, Fore. Error =γ +γ ·Fore. Error +γ ·lnSales +α +D +η (2) it 0 1 i,t−1 2 i,t−1 i st it whereFore. Error =lnSales −lnE [Sales ]denotesthediscrepancybetweencurrentsalesand it it t−1 it firm i forecast for time t. γ is our coefficient of interest; we expect γ < 0 as firms adjust their 1 1 forecast in response to past errors. We introduced past sales to absorb productivity shocks and the effect of market conditions. Among other controls, we consider the role of four factors: experience, global conditions, firm characteristics, and other forecast errors. As in the previous section, we proxy experience using the share of sales in the market and firm age; here, we also introduce the number of years a firm has been exporting to directly identify firm experience in foreign markets. A second set of indicators capturestheroleofglobalmarketconditionsandincludespastsalesinothermarkets,pastprojection errors, and price changes. The influence of firm characteristics is outlined by employment. A last group of regressors includes the errors in the forecast of other key variables–namely, employment and prices–which we expect to propagate through the sales projection. Finally, firmfixedeffectandsector-timedummiesextracttime-invariantfirmcharacteristicsand sector-time-specific shocks. Results Tables 10 and 11 present the estimates of model (2) for the domestic and foreign markets, respectively. Similaritiesbetweenthetwomodelsarelimitedtotheroleofforecasterrorsandpastsales. As expected, pasterrorsare negativelycorrelatedwith revisionstocurrent projections, suggestingthat firms tend to improve their forecasts by learning from previous errors. Comparing these estimates with the results in tables A2 and A3, where the forecast error is defined as the absolute deviation between realized sales and the projection, we document that both the size and the sign of the error matter in the domestic market, while the absolute deviation of the outcome from the projection outweighstheeffectofpositivevs. negativedeviationsinforeignmarkets. Usingthecoefficientfrom 9

column (6) in tables 10 and 11, we find that a 1-percent increase in past projection errors implies a 0.22-percent lower revision to the current forecast in the domestic and foreign markets. Similarlytopasterrors,pastsalesarenegativelycorrelatedwithprojectionerrors;lookingatthe effect of past shipments, one-standard-deviation higher sales are associated with 21 percent of a sd lowerforecastmissinthedomesticmarketandwith90percentofasdlowererrorinforeignmarkets. Thenegativecorrelationbetweensalesandtheforecasterrormightcaptureaproductivityadvantage of large firms over small firms; a similar advantage across potential exporters is documented by Dickstein and Morales [2015]. Table 10 shows that most other correlates do not significantly affect the projection errors in the domestic market; only errors to the forecast of other key variables significantly contribute to explaining the discrepancy between sales and the projection: errors in predicting employment and future price changes tend to propagate, magnifying the difference between current and expected sales. The results for foreign markets are more articulated (table 11). In particular, our proxies of experience continue to carry diverging signals. While age and the number of years a firm has operated abroad are negatively related to the size of the projection error, firms tend to make bigger errors in forecasting foreign shipments the larger the export share of sales. The global contour is among the factors inversely related to the forecast accuracy: Larger sales in the home market and larger price changes are positively correlated with revisions to the forecast, suggesting that firms do not fully incorporate changes to the near-term outlook into sales expectations. Finally, as for the domestic market, errors to employment feed into the foreign sales forecast, magnifying the magnitude of the errors. 5 Conclusions Thispaperexploresfirmforecastingstrategies. ExploitinganoveldatasetonItalianmanufacturing, we analyze how firms formulate their forecast and the accuracy of their predictions. Looking both across firms and over time, we find that current sales tend to explain a large share of the variability of the forecast. The foreign market forecast also benefits from the knowledge of domestic market conditions, consistent with the findings of Dickstein and Morales [2015], which fail to reject that 10

potential exporters know their own domestic sales in the previous year. In our analysis, we control for aggregate exports to each markets and the distance to each destination, other variables that DicksteinandMorales[2015]cannotrejectfromtheinformationsetofpotentialexporters,byadding sector-time dummies and firm fixed effects to our models. Moreover, our results point to the effect of experience and market-specific characteristics in shaping the projection. In terms of accuracy, we documentthatlargerfirmshaveanadvantageoversmallerfirmsincorrectlypredictingsales. Among other factors, while past errors are the only predictor that improves the home-market projection, experience and global condition also play a role in the foreign market specification. Taking into accountthoseadditionalfactorsiskeytoobtainingmorepreciseestimatesofstructuralparameters. 11

Figure 1: Comparison between Current Sales and Forecast, Domestic Market, 1995–2015 Figure 2: Comparison between Current Sales and Forecast, Foreign Markets, 1995–2015 12

Figure 3: Comparison between Current Sales and Past Forecast, Domestic Market, 1996–2015 Figure 4: Comparison between Current Sales and Past Forecast, Foreign Markets, 1996–2015 13

Table 1: Growth Rates Distribution in 2005: Domestic Market Current Expectationfor2006 p10 -24% -35% p25 -11% -13% p50 -1% 0% p75 +8% 14% p90 +22% 35% Avg -1% 0% Sd 22% 31% Note: Comparisonbetweenrealizedand expectedgrowthratesofdomesticsales. Table 2: Growth Rates Distribution in 2005: Foreign Markets Current Expectationfor2006 p10 -11% -10% p25 -2% -2% p50 1% 4% p75 7% 12% p90 18% 28% Avg 2% 6% Sd 28% 22% Note: Comparisonbetweenrealizedand expectedgrowthratesofforeignsales. 14

Table 3: Sales Forecast: Domestic Market (1) (2) (3) (4) (5) (6) (7) Variables lnEt[Dom. Salest+1] lnDom. Salest 0.989*** 0.996*** 0.993*** 0.994*** 0.992*** 1.024*** 1.016*** (0.002) (0.002) (0.002) (0.002) (0.002) (0.037) (0.038) Dom. ForecastErrort -0.167*** -0.197*** -0.197*** -0.198*** -0.227*** -0.224*** (0.025) (0.027) (0.027) (0.027) (0.038) (0.038) Dom. Sharet -0.026*** -0.006 0.002 0.039*** (0.007) (0.010) (0.010) (0.011) Age -0.007*** -0.007*** -0.008*** -0.008*** (0.002) (0.002) (0.003) (0.003) lnExports 0.002** 0.002 0.002 t (0.001) (0.004) (0.004) Exp. ForecastErrort -0.005*** -0.005 -0.004 (0.002) (0.005) (0.005) lnPriceChange 0.060 0.057 0.066* t (0.038) (0.038) (0.038) lnDom. Salest−1 -0.069** -0.072** (0.030) (0.030) lnExports -0.0002 0.001 t−1 (0.004) (0.004) lnEmpl 0.022*** t (0.004) Sector-Year n n y y y y y Obs. 24,500 24,500 24,500 24,500 24,500 24,500 24,500 R2 0.976 0.977 0.978 0.978 0.978 0.978 0.978 lnE [Dom. Sales ]: expected domestic sales at t+1 (in log-s). t t+1 lnDom. Sales : domestic sales at t (in log-s). t Dom. Forecast Error : forecast error in domestic sales at t, measured as the difference between ext pected and actual domestic sales at t (in log-s). Dom. Share : share of domestic out of total sales. t Age: years since birth, in log-s. lnExports : foreign sales at t (in log-s). t Exp. Forecast Error : forecast error in exports at t, measured as the difference between expected and t actual foreign sales at t (in log-s). lnPrice Change : change in price at t compared to the level of prices reported by the firm for t−1 (in t log). lnDom. Sales : domestic sales at t−1 (in log-s). t−1 lnExports : foreign sales at t−1 (in log-s). t−1 lnEmpl : total employment at t (in log-s). t Legend: ∗∗∗ significant at 1%, ∗∗ at 5%, ∗ at 10%. Notes: Firm-level regressions, 1996-2015. Specifications (4)-(6) also include sales and exports at t−2; those coefficients are omitted as they are not statistically significant. Standard errors are clustered at the firm level. 15

Table 4: Sales Forecast: Foreign Markets (1) (2) (3) (4) (5) (6) (7) (cid:2) (cid:3) Variables lnEt Exports t+1 lnExports 0.992*** 0.995*** 0.994*** 0.994*** 0.993*** 1.032*** 1.032*** t (0.001) (0.001) (0.001) (0.002) (0.003) (0.023) (0.023) Exp. ForecastErrort -0.060*** -0.060*** -0.060*** -0.061*** -0.096*** -0.095*** (0.008) (0.008) (0.008) (0.008) (0.022) (0.022) Exp. Sharet 0.010 0.019 0.035 0.013 (0.021) (0.028) (0.027) (0.029) Age 0.004 0.003 0.005 0.005 (0.007) (0.007) (0.007) (0.007) lnDom. Salest 0.003 0.050* 0.045 (0.004) (0.028) (0.028) Dom. ForecastErrort -0.026* -0.061** -0.059** (0.014) (0.029) (0.029) lnPriceChange -0.027 -0.054 -0.049 t (0.119) (0.116) (0.116) lnExports -0.043* -0.043* t−1 (0.023) (0.023) lnDom. Salest−1 -0.029 -0.031 (0.025) (0.025) lnEmpl 0.013 t (0.008) Sector-Year n n y y y y y Obs. 24,500 24,500 24,500 24,500 24,500 24,500 24,500 R2 0.975 0.975 0.975 0.975 0.975 0.975 0.975 (cid:2) (cid:3) lnE Exports : expected foreign sales at t+1 (in log-s). t t+1 lnExports : foreign sales at t (in log-s). t Exp. Forecast Error : forecast error in exports at t, measured as the difference between expected and t actual foreign sales at t (in log-s). Exp. Share : share of foreign out of total sales. t Age: years since birth, in log-s. lnDom. Sales : domestic sales at t (in log-s). t Dom. Forecast Error : forecast error in domestic sales at t, measured as the difference between ext pected and actual domestic sales at t (in log-s). lnPrice Change : change in price at t compared to the level of prices reported by the firm for t−1 (in t log). lnDom. Sales : domestic sales at t−1 (in log-s). t−1 lnExports : foreign salesat t−1 (in log-s). t−1 lnEmpl : total employment at t (in log-s). t Legend: ∗∗∗ significant at 1%, ∗∗ at 5%, ∗ at 10%. Notes: Firm-level regressions, 1996-2015. Specifications (4)-(6) also include sales and exports at t−2; those coefficients are omitted as they are not statistically significant. Standard errors are clustered at the firm level. 16

Table 5: Sales Forecast: Domestic Market (1) (2) (3) (4) (5) (6) (7) Variables lnEt[Dom. Salest+1] lnDom. Salest 0.856*** 0.881*** 0.859*** 0.853*** 0.847*** 0.764*** 0.754*** (0.019) (0.014) (0.018) (0.019) (0.019) (0.032) (0.033) Dom. ForecastErrort -0.069*** -0.081*** -0.077*** -0.075*** -0.003 0.0001 (0.021) (0.023) (0.022) (0.022) (0.031) (0.031) Dom. Sharet 0.062* 0.110*** 0.062 0.099** (0.037) (0.041) (0.040) (0.043) Age -0.014 -0.014 -0.015 -0.016 (0.010) (0.010) (0.010) (0.010) lnExports 0.009*** 0.004 0.003 t (0.002) (0.004) (0.003) Exp. ForecastErrort -0.007*** -0.005 -0.004 (0.002) (0.003) (0.003) lnPriceChange 0.045 0.071** 0.072** t (0.028) (0.034) (0.033) lnDom. Salest−1 0.075*** 0.068*** (0.023) (0.023) lnExports 0.001 0.001 t−1 (0.003) (0.003) lnEmpl 0.075*** t (0.019) FirmFE y y y y y y y Sector-Year n n y y y y y Obs. 24,500 24,500 24,500 24,500 24,500 24,500 24,500 R2 0.735 0.738 0.747 0.748 0.748 0.750 0.751 lnE [Dom. Sales ]: expected domestic sales at t+1 (in log-s). t t+1 lnDom. Sales : domestic sales at t (in log-s). t Dom. Forecast Error : forecast error in domestic sales at t, measured as the difference between ext pected and actual domestic sales at t (in log-s). Dom. Share : share of domestic out of total sales. t Age: years since birth, in log-s. lnExports : foreign sales at t (in log-s). t Exp. Forecast Error : forecast error in exports at t, measured as the difference between expected t and actual foreign sales at t (in log-s). lnPrice Change : change in price at t compared to the level of prices reported by the firm for t−1 t (in log). lnDom. Sales : domestic sales at t−1 (in log-s). t−1 lnExports : foreign salesat t−1 (in log-s). t−1 lnEmpl : total employment at t (in log-s). t Legend: ∗∗∗ significant at 1%, ∗∗ at 5%, ∗ at 10%. Notes: FE Firm-level regressions, 1996-2015. Specifications (4)-(6) also include sales and exports at t−2. Standard errors are clustered at the firm level. 17

Table 6: Sales Forecast: Foreign Markets (1) (2) (3) (4) (5) (6) (7) (cid:2) (cid:3) Variables lnEt Exports t+1 lnExports 0.926*** 0.931*** 0.928*** 0.922*** 0.918*** 0.830*** 0.829*** t (0.009) (0.009) (0.010) (0.011) (0.011) (0.026) (0.025) Exp. ForecastErrort -0.009 -0.008 -0.006 -0.005 0.080*** 0.081*** (0.009) (0.009) (0.009) (0.009) (0.024) (0.023) Exp. Sharet 0.175** 0.305*** 0.132 0.098 (0.084) (0.104) (0.106) (0.103) Age -0.029 -0.032 -0.034 -0.034 (0.020) (0.020) (0.021) (0.021) lnDom. Salest 0.069*** 0.073** 0.063* (0.022) (0.037) (0.037) Dom. ForecastErrort -0.063*** -0.071** -0.069** (0.020) (0.033) (0.033) lnPriceChange -0.259*** -0.225** -0.224** t (0.099) (0.096) (0.096) lnExports 0.104*** 0.105*** t−1 (0.025) (0.025) lnDom. Salest−1 -0.033 -0.039 (0.028) (0.028) lnEmpl 0.068* t (0.040) FirmFE y y y y y y y Sector-Year n n y y y y y Obs. 24,500 24,500 24,500 24,500 24,500 24,500 24,500 R2 0.790 0.790 0.793 0.793 0.793 0.795 0.796 (cid:2) (cid:3) lnE Exports : expected foreign sales at t+1 (in log-s). t t+1 lnExports : foreign sales at t (in log-s). t Exp. Forecast Error : forecast error in exports at t, measured as the difference between expected t and actual foreign sales at t (in log-s). Exp. Share : share of foreign out of total sales. t Age: years since birth, in log-s. lnDom. Sales : domestic sales at t (in log-s). t Dom. Forecast Error : forecast error in domestic sales at t, measured as the difference between t expected and actual domestic sales at t (in log-s). lnPrice Change : change in price at t compared to the level of prices reported by the firm for t−1 t (in log). lnDom. Sales : domestic sales at t−1 (in log-s). t−1 lnExports : foreign salesat t−1 (in log-s). t−1 lnEmpl : total employment at t (in log-s). t Legend: ∗∗∗ significant at 1%, ∗∗ at 5%, ∗ at 10%. Notes: FE Firm-level regressions, 1996-2015. Specifications (4)-(6) also include sales and exports at t−2; those coefficients are omitted as they are not statistically significant. Standard errors are clustered at the firm level. 18

Table 7: Foreign Sales Forecast: European Union (1) (2) (3) (4) (5) (6) Variables lnEt (cid:2) ExportsE t+ U 1 (cid:3) lnExportsEU 0.991*** 0.991*** 0.991*** 0.992*** 0.992*** 0.992*** t (0.005) (0.005) (0.005) (0.005) (0.005) (0.006) EUExp. Share 0.001 -0.002 -0.003 -0.004 (0.002) (0.004) (0.005) (0.005) lnExports -0.002 -0.008 -0.009 t (0.003) (0.009) (0.009) Exp. ForecastErrort -0.004 -0.002 -0.003 (0.004) (0.003) (0.003) lnExports 0.006 0.006 t−1 (0.008) (0.008) lnEmpl 0.002 t (0.007) lnPriceChange 0.004 t (0.042) Sector-Year n y y y y y Obs. 885 885 885 885 885 885 R2 0.978 0.978 0.978 0.978 0.978 0.978 lnE (cid:2) ExportsEU(cid:3) : expected foreign sales in EU at t+1 (in log-s). t t+1 lnExportsEU: foreign sales in EU at t (in log-s). t EU Exp. Share : share of EU exports out of foreign sales. t lnExports : foreign sales at t (in log-s). t Exp. Forecast Error : forecast error in exports at t, measured as the difference bet tween expected and actual foreign sales at t (in log-s). lnExports : foreign salesat t−1 (in log-s). t−1 lnEmpl : total employment at t (in log-s). t lnPrice Change : change in price at t compared to the level of prices reported by the t firm for t−1 (in log). Legend: ∗∗∗ significant at 1%, ∗∗ at 5%, ∗ at 10%. Notes: Firm-level regressions, 2008. Standard errors are clustered at the firm level. 19

Table 8: Foreign Sales Forecast: North America (1) (2) (3) (4) (5) (6) Variables lnEt (cid:2) ExportsN t+ A 1 (cid:3) lnExportsNA 0.972*** 0.973*** 0.972*** 0.975*** 0.975*** 0.975*** t (0.009) (0.009) (0.010) (0.010) (0.010) (0.010) NAExp. Share 0.067 0.035 0.005 0.047 (0.046) (0.061) (0.065) (0.068) lnExports -0.006 -0.043 -0.031 t (0.006) (0.024) (0.024) Exp. ForecastErrort 0.030** 0.042** 0.042** (0.014) (0.019) (0.018) lnExports 0.037 0.036 t−1 (0.023) (0.023) lnEmpl -0.031*** t (0.012) lnPriceChange -0.181** t (0.072) Sector-Year n y y y y y Obs. 885 885 885 885 885 885 R2 0.946 0.947 0.947 0.947 0.947 0.948 lnE (cid:2) ExportsNA(cid:3) : expected foreign sales in North America at t+1 (in log-s). t t+1 lnExportsNA: foreign sales in North America at t (in log-s). t NA Exp. Share : share of North American exports out of foreign sales. t lnExports : foreign sales at t (in log-s). t Exp. Forecast Error : forecast error in exports at t, measured as the difference between t expected and actual foreign sales at t (in log-s). lnExports : foreign salesat t−1 (in log-s). t−1 lnEmpl : total employment at t (in log-s). t lnPrice Change : change in price at t compared to the level of prices reported by the t firm for t−1 (in log). Legend: ∗∗∗ significant at 1%, ∗∗ at 5%, ∗ at 10%. Notes: Firm-level regressions, 2008. Standard errors are clustered at the firm level. 20

Table 9: Foreign Sales Forecast: China (1) (2) (3) (4) (5) (6) Variables lnEt (cid:2) ExportsC t+ h 1 ina(cid:3) lnExportsChina 0.986*** 0.984*** 0.981*** 0.974*** 0.974*** 0.974*** t (0.012) (0.013) (0.014) (0.014) (0.014) (0.014) ChinaExp. Share 2.146 3.492 3.507 3.334 (2.212) (2.421) (2.433) (2.462) lnExports 0.012** 0.006 0.001 t (0.005) (0.008) (0.008) Exp. ForecastErrort -0.001 0.001 0.0005 (0.002) (0.003) (0.003) lnExports 0.006 0.005 t−1 (0.006) (0.006) lnEmpl 0.018 t (0.011) lnPriceChange -0.004 t (0.069) Sector-Year n y y y y y Obs. 885 885 885 885 885 885 R2 0.922 0.923 0.923 0.924 0.924 0.924 lnE (cid:2) ExportsChina(cid:3) : expected foreign sales in China at t+1 (in log-s). t t+1 lnExportsChina: foreign sales in China at t (in log-s). t China Exp. Share : share of Chinese exports out of foreign sales. t lnExports : foreign sales at t (in log-s). t Exp. Forecast Error : forecast error in exports at t, measured as the difference bet tween expected and actual foreign sales at t (in log-s). lnExports : foreign salesat t−1 (in log-s). t−1 lnEmpl : total employment at t (in log-s). t lnPrice Change : change in price at t compared to the level of prices reported by the t firm for t−1 (in log). Legend: ∗∗∗ significant at 1%, ∗∗ at 5%, ∗ at 10%. Notes: Firm-level regressions, 2008. Standard errors are clustered at the firm level. 21

Table 10: Sales Forecast Error: Domestic Market (1) (2) (3) (4) (5) (6) Variables lnDomSalest−lnEt−1[DomSalest] Fore. Errort−1 -0.234*** -0.216*** -0.216*** -0.217*** -0.217*** -0.217*** (0.029) (0.032) (0.032) (0.032) (0.032) (0.032) lnDomSalest−1 -0.048** -0.058** -0.055** -0.054* -0.053* (0.020) (0.023) (0.024) (0.028) (0.028) DomSharet−1 0.077 0.059 0.058 0.049 (0.069) (0.083) (0.088) (0.088) Age -0.014 -0.014 -0.014 -0.012 (0.016) (0.016) (0.016) (0.016) FMktExperience -0.040 -0.040 -0.040 -0.038 (0.026) (0.026) (0.026) (0.025) ForeExportErrort−1 -0.002 -0.002 -0.002 (0.003) (0.003) (0.003) lnExports -0.002 -0.002 -0.002 t−1 (0.005) (0.005) (0.005) lnPriceChange -0.072 -0.072 -0.047 t−1 (0.079) (0.079) (0.084) lnEmpl -0.003 0.003 t−1 (0.027) (0.027) ForeEmplErrort 0.213*** (0.046) ForePriceErrort 0.286*** (0.066) FirmFE y y y y y y Sector-Year n y y y y y Obs. 17,613 17,613 17,613 17,613 17,613 17,613 R2 0.183 0.185 0.185 0.186 0.186 0.191 lnDom Sales −lnE [Dom Sales ]: forecast error in domestic sales at t (in log-s). t t−1 t Fore. Error : forecast error at t−1 in domestic sales. t−1 lnDom. Sales : domestic sales at t−1. t−1 Dom. Share : share of domestic out of total sales. t−1 Age: years since birth, in log-s. F Mkt Experience: years active in the foreign markets. Fore. Export Error : forecast error at t−1 in foreign sales. t−1 lnExports : foreign sales at t−1 (in log-s). t−1 lnPrice Change : change in prices at t−1 (in log). t−1 lnEmpl : total employment at t−1 (in log-s). t−1 Fore Empl Error : forecast error in employment at t. t Fore Price Error : forecast error in prices at t. t Legend: ∗∗∗ significant at 1%, ∗∗ at 5%, ∗ at 10%. Notes: Firm FE regressions, 1996-2015. Standard errors are clustered at the firm level. 22

Table 11: Sales Forecast Error: Foreign Markets (1) (2) (3) (4) (5) (6) Variables lnExports t −lnEt−1[Exports t ] ForeExportErrort−1 -0.276*** -0.225*** -0.223*** -0.219*** -0.218*** -0.218*** (0.018) (0.021) (0.021) (0.021) (0.022) (0.022) lnExports -0.250*** -0.295*** -0.338*** -0.339*** -0.340*** t−1 (0.037) (0.040) (0.045) (0.045) (0.045) Exp. Sharet−1 0.815*** 1.544*** 1.514*** 1.530*** (0.162) (0.255) (0.251) (0.250) Age -0.050** -0.051** -0.052** -0.048** (0.025) (0.024) (0.024) (0.024) FMktExperience -0.018 -0.015 -0.015 -0.015 (0.060) (0.061) (0.061) (0.061) ForeErrort−1 0.015 0.018 0.019 (0.024) (0.023) (0.023) lnDomSalest−1 0.225*** 0.212*** 0.213*** (0.047) (0.046) (0.046) lnPriceChange 0.285 0.291 0.319* t−1 (0.177) (0.177) (0.177) lnEmpl 0.049 0.058 t−1 (0.060) (0.060) ForeEmplErrort 0.387*** (0.113) ForePriceErrort 0.270 (0.215) FirmFE y y y y y y Sector-Year n y y y y y Obs. 17,613 17,613 17,613 17,613 17,613 17,613 R2 0.139 0.181 0.186 0.191 0.191 0.193 lnExports −lnE [Exports ]: forecast error in exports at t (in log-s). t t−1 t Fore. Export Error : forecast error at t−1 in foreign sales. t−1 Exp. Share : share of exports out of total sales. t−1 Age: years since birth, in log-s. F Mkt Experience: years active in the foreign markets. lnExports : foreign sales at t−1 (in log-s). t−1 Fore. Error : forecast error at t−1 in domestic sales. t−1 lnDom. Sales : domestic sales at t−1. t−1 lnPrice Change : change in prices at t−1 (in log). t−1 lnEmpl : total employment at t−1 (in log-s). t−1 Fore Empl Error : forecast error in employment at t. t Fore Price Error : forecast error in prices at t. t Legend: ∗∗∗ significant at 1%, ∗∗ at 5%, ∗ at 10%. Notes: Firm FE regressions, 1996-2015. Standard errors are clustered at the firm level. 23

References Banca d’Italia. Indagine sulle Imprese Industriali e dei Servizi, 1995-2015. Facundo Albornoz, H´ector F Calvo Pardo, Gregory Corcos, and Emanuel Ornelas. Sequential Exporting. Journal of International Economics, 88(1):17–31, 2012. Henry Anderson. Problems Peculiar to Export Sales Forecasting. The Journal of Marketing, 24(4): 39–42, 1960. Ru¨digerBachmann,SteffenElstner,andEricRSims. UncertaintyandEconomicActivity: Evidence from Business Survey Data. American Economic Journal: Macroeconomics, 5(2):217–249, 2013. STamerCavusgil.DifferencesamongExportingFirmsbasedonTheirDegreeofInternationalization. Journal of Business Research, 12(2):195–208, 1984. Cheng Chen, Tatsuro Senga, Chang Sun, and Hongyong Zhang. Policy Uncertainty and Foreign Direct Investment: Evidence from the China-Japan Island Dispute. 2016. Flavio Cunha and James J Heckman. Identifying and Estimating the Distributions of Ex-Post and Ex-Ante Returns to Schooling. Labour Economics, 14(6):870–893, 2007. Douglas J Dalrymple. Sales Forecasting Practices: Results from a United States Survey. International Journal of Forecasting, 3(3-4):379–391, 1987. AdamantiosDiamantopoulosandHeidiWinklhofer. TheImpactofFirmandExportCharacteristics on the Accuracy of Export Sales Forecasts: Evidence from UK Exporters. International Journal of Forecasting, 15(1):67–81, 1999. Michael J Dickstein and Eduardo Morales. What do Exporters Know? Technical Report 21351, National Bureau of Economic Research, 2015. Charles F Manski. Adolescent Econometricians: How do Youth Infer the Returns to Schooling? In Studies of Supply and Demand in Higher Education, pages 43–60. University of Chicago Press, 1993. Charles F Manski. Measuring Expectations. Econometrica, 72(5):1329–1376, 2004. Maria D. Tito. Exporting and Skilled Demand: Evidence from Shocks to Expectations. 2015. 24

Heidi Winklhofer, Adamantios Diamantopoulos, and Stephen F Witt. Forecasting Practice: A Review of the Empirical Literature and an Agenda for Future Research. International Journal of Forecasting, 12(2):193–221, 1996. 25

A Additional Tables Table A1: Summary Statistics Variable Mean Std. Dev. Domestic Salesa 9.42 1.61 Foreign Salesa 7.18 4.00 Expectations of Domestic Salesa 9.44 1.61 Expectations of Foreign Salesa 7.23 4.02 Domestic Forecast Error -0.02 0.41 Foreign Forecast Error -0.01 1.49 aValues are in (log) thousand euros. Domestic Forecast Error: difference between log-sales and log-expectations in domestic sales. Foreign Forecast Error: difference between log-sales and log-expectations in foreign sales. 26

Table A2: Sales Forecast Error: Domestic Market (1) (2) (3) (4) (5) (6) Variables |lnDomSalest−lnEt−1[DomSalest]| AbsFore. Errort−1 -0.102** -0.079 -0.079 -0.083 -0.082 -0.083 (0.046) (0.051) (0.051) (0.052) (0.052) (0.052) lnDomSalest−1 -0.063*** -0.057*** -0.060*** -0.061** -0.060** (0.018) (0.020) (0.023) (0.027) (0.027) DomSharet−1 -0.047 -0.024 -0.022 -0.026 (0.057) (0.077) (0.082) (0.082) Age 0.006 0.005 0.005 0.006 (0.012) (0.012) (0.012) (0.012) FMktExperience -0.018 -0.018 -0.018 -0.019 (0.019) (0.019) (0.019) (0.019) AbsForeExportErrort−1 -0.009** -0.009** -0.009** (0.004) (0.004) (0.004) lnExports 0.006 0.006 0.005 t−1 (0.005) (0.005) (0.005) lnPriceChange 0.066 0.067 0.066 t−1 (0.064) (0.064) (0.065) lnEmpl 0.003 0.004 t−1 (0.025) (0.025) AbsForeEmplErrort 0.189*** (0.033) AbsForePriceErrort 0.059 (0.073) FirmFE y y y y y y Sector-Year n y y y y y Obs. 17,613 17,613 17,613 17,613 17,613 17,613 R2 0.143 0.150 0.150 0.151 0.151 0.155 |lnDom Sales −lnE [Dom Sales ]|: absolute deviation between domestic sales and expect t−1 t tation at t (in log-s). Abs Fore. Error : absolute deviation between domestic sales and expectation at t−1. t−1 lnDom. Sales : domestic sales at t−1. t−1 Dom. Share : share of domestic out of total sales. t−1 Age: years since birth, in log-s. F Mkt Experience: years active in the foreign markets. Abs Fore. Export Error : absolute deviation between foreign sales and expectation at t−1. t−1 lnExports : foreign sales at t−1 (in log-s). t−1 lnPrice Change : change in prices at t−1 (in log). t−1 lnEmpl : total employment at t−1 (in log-s). t−1 Abs Fore Empl Error : absolute deviation between employment and expectations at t. t Abs Fore Price Error : absolute deviation between prices changes and expectation at t. t Legend: ∗∗∗ significant at 1%, ∗∗ at 5%, ∗ at 10%. Notes: Firm FE regressions, 1996-2015. Standard errors are clustered at the firm level. 27

Table A3: Sales Forecast Error: Foreign Markets (1) (2) (3) (4) (5) (6) Variables |lnExports t −lnEt−1[Exports t ]| AbsForeExportErrort−1 -0.219*** -0.161*** -0.159*** -0.157*** -0.157*** -0.157*** (0.017) (0.019) (0.019) (0.019) (0.019) (0.019) lnExports -0.285*** -0.327*** -0.372*** -0.373*** -0.373*** t−1 (0.028) (0.031) (0.035) (0.035) (0.035) Exp. Sharet−1 0.760*** 1.522*** 1.491*** 1.494*** (0.134) (0.210) (0.207) (0.207) Age -0.036 -0.037* -0.037* -0.036* (0.022) (0.022) (0.022) (0.022) FMktExperience -0.122** -0.119** -0.119** -0.118* (0.061) (0.060) (0.060) (0.061) AbsForeErrort−1 -0.036* -0.033 -0.033 (0.021) (0.021) (0.021) lnDomSalest−1 0.253*** 0.239*** 0.239*** (0.039) (0.038) (0.039) lnPriceChange 0.370*** 0.376*** 0.377*** t−1 (0.141) (0.141) (0.143) lnEmpl 0.050 0.051 t−1 (0.050) (0.050) AbsForeEmplErrort 0.288** (0.131) AbsForePriceErrort 0.356* (0.212) FirmFE y y y y y y Sector-Year n y y y y y Obs. 17,613 17,613 17,613 17,613 17,613 17,613 R2 0.117 0.189 0.195 0.203 0.203 0.204 |lnExports −lnE [Exports ]|: absolute deviation between exports and expectations at t (in t t−1 t log-s). Abs Fore. Export Error : absolute deviation between foreign sales and expectations at t−1. t−1 Exp. Share : share of exports out of total sales. t−1 Age: years since birth, in log-s. F Mkt Experience: years active in the foreign markets. lnExports : foreign sales at t−1 (in log-s). t−1 Abs Fore. Error : absolute deviation between domestic sales and expectations at t−1. t−1 lnDom. Sales : domestic sales at t−1. t−1 lnPrice Change : change in prices at t−1 (in log). t−1 lnEmpl : total employment at t−1 (in log-s). t−1 Abs Fore Empl Error : absolute deviation between employment and expectations at t. t Abs Fore Price Error : absolute deviation between prices and expectations at t. t Legend: ∗∗∗ significant at 1%, ∗∗ at 5%, ∗ at 10%. Notes: Firm FE regressions, 1996-2015. Standard errors are clustered at the firm level. 28