PRELIMINARY VERSION Tax Incentives to Encourage Corporate Investment in Latvia1 Mihails HAZANS2 (University of Latvia, IZA, and GLO) and Anna Pluta (BICEPS) This paper analyzes the effect of accelerated depreciation (AD) policy on firm investment using administrative firm-level data for 2007–14. We find that past use of AD had a positive effect on firm investment rates. The effect is stronger in industries with most of their assets in long- duration categories, among enterprises with fewer than six employees and among firms whose turnover is highly volatile. AD of new equipment has a significant effect both on next year investment rate and on probability to invest next year, but only for firms with six to ten workers. AD in specially assisted areas has boosted investment decisions after lifting the restrictive project requirement for this category of AD. 1. Introduction During 2006-2017, Latvia has applied a rather generous accelerated depreciation (hereafter AD) policy to stimulate firm investment. Neighboring Estonia has applied, since 2000, a zero corporate income tax on reinvested profit. During 2008-2015, business investment rate (defined as gross investment/ gross value added of non-financial corporations) in Latvia was in line with other Baltic countries and Poland. However, since 2010, Latvia lags behind Estonia in terms of investment per person employed, and the gap is increasing (see Annex 1 for details). Moreover, in Latvia economic activities with high investment per person employed and high business investment rate are not among those with the highest apparent labor productivity. This might indicate that the AD scheme in Latvia was too general and/or too generous, resulting in over-investment in less productive firms or industries. The purpose of this paper is to shed light on responses to the following questions: 1. Did the AD policy encourage firms to invest in Latvia? 2. How did past experience in using AD (in terms of relative and absolute size of accelerated depreciation value and the magnitude of the reduction of taxable income) affect the size of investment? 3. How do the effects of AD of new equipment, AD in SAAs, and other types of AD compare? 1 The paper has benefited from many useful discussions with Emilia Skrok, Irem Guceri, Palma Mosberger, and three anonimous referees. The usual disclaimer applies. 2 Corresponding author e-mail: mihazan@lanet.lv. 1 4. Did lifting the project requirement for AD in SAAs have a positive effect on investment? Previous studies have identified several channels through which accelerated depreciation (AD) might affect investments: (1) It lowers the user cost of capital (Jorgenson 1963; Zwick and Mahon 2017). (2) For firms on a tight margin it relaxes the cash flow constraint (Kaplan and Zingales 1997, Stein 2003, Devereux and Liu 2017, Zwick and Mahon 2017, among others). (3) Managers keen on tax saving might use AD for this reason but only when the tax benefits are immediate (Zwick and Mahon 2017)—as is often the case under the Latvian AD policy. We are also going to explore the uncertainty-mitigating effect of accelerated depreciation: firms can afford more risky investment if the cash flow from AD is sufficient for "business-as-usual" investment as a plan B. In what follows, Section 2 details Latvia’s AD policy and compares it to U.S. bonus depreciation, which has been intensively studied (by among others House and Shapiro 2008, Zwick and Mahon 2017, Ohrn forthcoming). Section 3 summarizes the approaches and main findings of previous studies on temporary and permanent investment tax incentives in various countries. Section 4 describes our data sources and the matching process we apply to merge corporate tax return data with the anonymized extracts from annual enterprise reports. Section 5 outlines the econometric methodology, describing the main variables and treatment and control groups. Section 6 presents the results. Section 7 draws conclusions. 2. Policy Context Since 2006 Latvia has used AD policy to stimulate firm investment. It allowed firms to use for tax purposes a larger asset depreciation value3 (ADV) than for balance sheet depreciation,4 so that firms were able to reduce taxable income by the difference between ADV and the balance sheet depreciation value (BDV). The AD policy has three main components: (1) the general AD scheme, (2) incentives to acquire new technological equipment, and (3) Investment incentives in SAAs. Closely related to AD (although not limited to it) are R&D incentives (effectively, the fourth component of AD policy). The general AD scheme sets out five asset categories, and the baseline depreciation rates range from 5 to 20 percent (Table 1). Because these rates are then doubled to determine the amount of depreciation for tax purposes, for most types of assets5 (Table 1) the effective rates of depreciation are twice the baseline. There are two main exceptions to the “double rate” rule: 3 We use the term “depreciation value” as synonymous with “depreciation expenses,” “depreciation amount,” or "depreciation deductions." 4 For balance sheet depreciation, firms can choose common methods, like straight line or double straight line. 5 Among assets to which depreciation does not apply are land, works of art and antiques, jewelry and other fixed assets that are not subject to physical or economic depreciation; and investment properties, organic assets, and long-term investments held for sale, which the taxpayer has chosen to value at their true value. 2 1. For passenger cars, motorcycles, sea and river, and air means of transport, a coefficient of 1.5 rather than 2 applies to the baseline rate of 20%, so that the AD rate is 30%. 2. Representation passenger cars (value above €50,000 without VAT) are not eligible for AD at all. The incentives for new technological equipment were applied to new production equipment acquired or established by the taxpayer in a taxation period commencing in 2006 or later and used in economic activities. For such assets, the acquisition or creation value was multiplied by 1.5 before calculating the depreciation deduction (in 2007 the multiplier was 1.4 and in 2008 it was 1.3). On top of this, the general AD scheme with double depreciation rates applied (Table 1). Table 1. Accelerated Depreciation and R&D Deductions, Latvia, 2007-17 Multipliers (coefficients) applied to Depreciation or asset value or R&D costs before deduction rate depreciation or deduction for tax purposes New equipment, SAAs (Specially Assets Baseline Double patents, trademarks a Assisted Areas) b 1. Buildings, structures, perennial plants 5% 10% No 1.5 2. Railway rolling stock and technological equipment, sea and river fleet vessels, fleet and 10% 20% 1.5 1.3 c port technological equipment, power equipment 3. Computing devices and related equipment, including printing devices, information systems, software products, data storage equipment, means 35 % 70% 1.5 1.8 of communication, copiers, and related equipment. Computer programs 4. Other fixed assets, except those in category 5 20 % 40% d 1.5 2.0 e 5. Oil exploration and extraction platforms, oil 7.5 % 15% 1.5 1.0 exploration and extraction ships 6. Patents, trademarks (excl. those in category 7) Straight line, 1.5 g 1.0 5 years R&D costs R&D costs 7. R&D costs f related to taxpayer's economic activity 100% No Before January 1, 2014, 1.0; after, 3.0 Source: Law on corporate income tax (as of 2017). Notes: a In 2007, the coefficient for new equipment was 1.4 and in 2008 it was 1.3. b If an asset falls in both the "New equipment" and "SAAs" categories, only one multiplier can be applied. SAAs were removed from the CIT law after January 1, 2014, but firms previously operating in SAAs could continue to apply the AD rules for assets bought before that date. c Except for railway rolling stock and marine and river fleet vehicles. d 30% for cars, motorcycles, sea and river vehicles, and aircraft. e Except for cars, motorcycles, sea and river vehicles, and aircraft. f Except for costs of geological exploration. g Starting in 2009. 3 Investment incentives in the specially assisted areas (SAAs) applied to taxpayers established and operating in such areas (the least developed municipalities in Latvia, listed in the Cabinet regulations amending the Regional Development Law, see Figure 1). Before calculating the depreciation value for fixed assets acquired by eligible firms and used for economic activities in the SAAs, asset acquisition or creation value could be multiplied by a coefficient varying from 1.3 to 2.0 depending on the category of assets (see the right-hand column of Table 1). R&D incentives allowed taxpayers to write off 100% (after January 1, 2014 – 300%) of the costs of research and development related to their economic activity (other than costs for geological exploration) in the year when the costs are incurred. Between 2006 and 2017, the Latvian AD policy was a permanent feature of the tax code, unlike U.S. bonus depreciation (see, e.g., House and Shapiro 2008, Zwick and Mahon 2017) but similar to the UK first-year allowance (see, e.g., Maffini et al. 2016). Due to multipliers, it is much more generous than bonus depreciation; see Table A2.1 in Annex 2). Taxpayers eligible for both the SSA incentives and the incentives to acquire new technological equipment could apply just one (at their own choice) to the same fixed asset. Up to 2010, the SSA incentives were project-based rather than universal: eligible taxpayers had to submit a project application to the Latvian Investment and Development Agency. In 2010, this requirement was lifted, but the list of SAAs was significantly reduced as of 2011. The data (see Table 2) suggest that the first change permanently increased the number of firms using this incentive. Lifting the project requirement allowed smaller firms in SAAs to benefit from AD, resulting in an immediate fall in median and average AD values in 2010. In 2011, these values declined even more dramatically, however (Table 2), plausibly, because medium-sized and less remote municipalities were excluded (Figure 1). 4 Figure 1. Specially Assisted Areas (SAA) in Latvia. Top: 2007–09. Bottom: 2011–14. Source: Latvian Ministry of Environment and Regional Development. Note: SAAs are shown in green. The list of SAAs for 2010 is similar to the one for 2007–09 but excludes three towns (Gulbene, Madona, and Kuldiga) and three smaller municipalities. 5 Table 2. Distribution of AD Value (in Thousands €) and Number of Firms Using AD in SAAs, 2008–14 Year p10 p25 p50 p75 p90 Mean N firms 2008 0.04 0.2 0.7 8.5 76.3 70.6 434 2009 1.40 3.8 32.8 188.0 312.3 175.0 91 2010 0.36 1.6 11.1 43.7 214.9 107.9 140 2011 0.10 0.4 2.2 9.5 40.0 29.2 1,186 2012 0.12 0.5 2.6 13.1 48.4 47.8 1,485 2013 0.12 0.5 2.4 12.3 44.7 25.6 1,516 2014 0.13 0.5 2.5 12.1 47.2 31.0 1,643 Source: MOF CIT data. Apparently, firms found the AD policy attractive. As shown in Table 3, AD deductions accounted for 11–12% of Latvia’s GDP in most years reviewed, though during the crisis it was about 13%. The AD policy apparently reduced firm taxable income by 2.8% of GDP in 2008, 1.5–1.8% in 2009– 11, and 2.4% in 2012–14. In 2009–14, more than 90% of all firms with depreciating assets applied AD schemes for tax purposes (Table 3). Indeed, in most cases AD makes it possible to write off an asset faster for tax purposes than for balance sheet needs, thus reducing taxable income. Table 3. Accelerated and Balance Sheet Depreciation Values, Latvian Firms, 2008–14 Accelerated Depreciation Balance Sheet ADV Coverage Year Value (ADV) Depreciation Value (BDV) ADV-BDV (ADV>0) % of Firms % of all # Firms # Firms with Firms in % of with € % of with BDV > CIT € Mill. GDP ADV>0 Mill. GDP BDV>0 % of GDP 0 data 2008 2,691 11.1 50,591 1999 8.2 57,075 2.8 88.6 66.1 2009 2,446 13.0 52,158 2109 11.2 55,511 1.8 94.0 67.6 2010 2,303 12.8 51,121 2030 11.3 54,954 1.5 93.0 63.7 2011 2,286 11.3 49,504 1953 9.6 53,317 1.6 92.8 60.4 2012 2,491 11.4 48,312 1960 9.0 51,502 2.4 93.8 57.2 2013 2,662 11.7 49,473 2,108 9.3 52,400 2.4 94.4 56.9 2014 2,684 11.4 49,216 2,120 9.0 51,908 2.4 94.8 56.0 Source: MOF CIT data. AD for new technological equipment accounted for only a small share (from 9% in 2009 to 14% in 2013) of the total AD value (Figure 2, left) but for a very substantial share of the total reduction in corporate taxable income (Figure 2, right). 6 Figure 2. AD Value and Implied Reduction in Corporate Taxable Income by Category, Latvia, 2008–14 3000 2500 2000 Million EUR 1500 1000 500 0 2008 2009 2010 2011 2012 2013 2014 2008 2009 2010 2011 2012 2013 2014 -500 Accelerated depreciation value Difference beween accelerated an balance sheet depreciation values New technology equipment Specially assisted territories Other Source: MOF CIT data. The share of new equipment in the reduction in corporate taxable income was 40–50% in 2008, 2009, 2011, and 2013; about 33% 2012 and 2014; and almost 100% in 2010 (Figure 2). Table A2.2 (Annex A) lists the top 25 sectors (by 2-digit NACE classification) in terms of AD-caused reduction of taxable corporate income in 2012–14; 15 of these 25 sectors belong to manufacturing, while other sectors using AD heavily include utilities, real estate, agriculture, forestry, and land transport. In manufacturing alone, the annual average reduction was €88 million (Table A2.2). By contrast, AD in the SAAs accounts for a very small share of total AD value and the total reduction in corporate taxable income. However, AD in the SAAs is still of utmost interest due to the change in eligibility conditions in 2010 and 2011, which makes it possible to identify the policy effect. As of 2018, Latvia has abandoned its AD policy because its tax reform introduced, among other changes, the Estonian model of a zero CIT rate on profit that is reinvested rather than distributed.6 3. Literature Review So far, only a few studies have investigated the effect of tax incentives and different types of depreciation, especially on business investment across sub-groups of firms. Bronzini et al. (2008) evaluated the impact of the investment tax credit on business investment in Italy. They focus on the tax credit, which is not restricted to profitable enterprises with tax liability but can also be deducted from any outstanding payment due to central government. The amount of tax credit differs by area of eligibility, and the amount of the deduction decreases as 6 See Jacobs et al (2017) for a discussion of possible advantages and disadvantages of applying the Estonian model in Latvia. 7 local development grows. The results suggest that the program has been effective in boosting investment. Devereux, Maffini, and Xing (2016) provide evidence of a substantial positive effect of higher depreciation allowances on firm investments. In the UK, firms that qualify as an SME can claim a higher first-year capital allowance than the larger firms if they were below two of three thresholds for turnover, total assets, and number of employees. In 2004, the UK more than doubled the turnover and total assets thresholds. The authors found that access to more generous capital allowances increases firm investment by 2.1–2.6 percentage points (pp) relative to firms that never qualified for the more generous treatment; at the mean, this is equivalent to an 11 percent increase in investment. Yagan (2015) studied the effect of the 2003 U.S. dividend tax cut on corporate investment and labor earnings. In his estimation, the tax cut caused zero change in corporate investment and employee compensation. Similarly, Desai and Goolsbee (2014) showed that the dividend tax cut, despite its high revenue cost, had minimal, if any, influence on investment incentives. Bonus depreciation, passed in the U.S. in 2002 (it expired at the end of 2004) and again in 2008 allowed firms to deduct from their taxable income a “bonus” percentage of the cost of investment purchases. House and Shapiro (2008) explored the effect of the bonus depreciation allowance in 2002–03. Only investment goods with a tax recovery period up to 20 years qualified. The results suggest that bonus depreciation had a powerful effect on the composition of investment, in that there were steep increases in capital investment in assets that benefited substantially from the policy. Recently Zwick and Mahon (2017) found that bonus depreciation had a substantial effect on investment in 2001–04 and 2008–10. Theirs was the most complete dataset yet applied to study U.S. business investment incentives, and their results suggest that the investment response is larger for small, cash-poor firms—but only when the policy generates immediate rather than future cash flows. Ohrn (forthcoming) estimated the response of manufacturing to bonus depreciation and depreciation allowances in the U.S. states that adopted such policies and found that both policies have been effective in boosting investment. The policies also affected employment and total production, but only several years after adoption. In a companion study, Ohrn (2018) examined how firms responded to the domestic production activities deduction, which allows firms to deduct a percentage of domestic manufacturing income from their taxable income and found that corporate tax rate reductions motivated larger firms with more cash flow to invest more, but smaller, more financially constrained firms were more responsive to depreciation policies. Edgerton (2010) found that tax incentives like bonus depreciation have the least impact on investment exactly when they are most likely to be used—during economic downturns when cash flows are low. To our best knowledge, this paper is the first to ask whether the investment tax incentives that Latvia put in place in 2007–12 were effective. (Forthcoming papers by Skrok et al. and Mosberger and Varga look at the effects of tax incentives in two other CEE countries, Poland and Hungary.) 8 4. Data 4.1. Sources of Firm Data The main data source for this paper (hereafter: CIT data) is the (anonymized) annual panel of CIT declarations of all 128,459 Latvian firms that paid CIT in 2008–14. In addition to pre-tax profit and loss statements, the data include detailed information necessary for calculating taxable income, such as total depreciation values for accounting purposes and AD values for tax purposes. For total depreciation value, the data report separately the depreciation of new equipment, assets employed in SAAs, and patents. The data include 4-digit NACE codes, 6-digit municipality codes, and type of settlement. The panel is not balanced, but for 47,280 firms data are available for each of the 7 years studies, and at least 5 years of data are available for more than 50% of all firms. However, the CIT data do not contain our key variables of interest, investments and fixed capital. As an additional data source we use anonymized extracts from annual reports of all Latvian enterprises7 for 2007–14, provided by Lursoft IT. For each firm and year, this dataset covers tangible fixed assets and intangibles at the end of the year, profit or loss before taxes, the 4-digit NACE code, registration year, legal form of the enterprise, turnover, number of persons employed, CIT paid for the given year, and a 4-digit municipality code. In merging the two datasets we used variables available in both (year, profit or loss, NACE, municipality, CIT paid). Doing so is complicated for several reasons, among them (1) the CIT declarations (extracted from the SRS data warehouse in 2016) include the most recent versions of profit data, which might differ from those in the annual reports.8 (2) Similarly, the NACE and the municipality codes for the same firm might differ because annual reports and CIT declarations are not submitted simultaneously. 4.2. The Matching Procedure Our matching procedure works as follows: Step 1a. Match by year profit or loss before taxation, 4-digit NACE code, and 4-digit municipality code. Step 1b. For every pair of firms matched in Step 1a for a given year, compare profit or loss before tax for other years. If the absolute difference does not exceed €10 in at least one other year, the two firms are considered fully matched. However, if a firm from CIT data can be matched in annual report data with more than one other firm (i.e., has multiple twins) it is not considered fully matched. Step 2a (for firms not fully matched): Match by year, profit or loss before taxation, 2-digit NACE code, and 4-digit municipality code. Step 2b: Similar to Step 1b, but follows the Step 2a result. Step 3a (for firms not yet fully matched): Match by year, profit or loss before taxation and 4-digit NACE code. Step 3b: Similar to Step 1b, but follows the Step 3a result. 7 Except for single-owner or family enterprises with turnover below the threshold making annual reports mandatory. 8 Previous studies have also pointed out the difficulties in reconciling tax return data with annual reports, see, e.g., Mills et al. 2002. 9 Step 4a (for firms still not yet fully matched): Match by year, profit or loss before taxation. and 2- digit NACE code. Step 4b: Similar to Step 1b, but responds to the Step 4a result. Step 5a (for any firms remaining unmatched): Match by year and profit or loss before taxation. Step 5b: Similar to Step 1b but responds to the Step 5a result. We use a three-stage procedure to match firms: After Steps 1–5, 67.5% of all enterprises in the State Revenue Service Database are matched. For the firms still unmatched, we implement a second and third round of the same five steps but this time allowing for a profit difference of up to €100 rather than €10.9 In the second and third round, only 1.5% of all firms from CIT data arematched. In the fourth round, we repeat the matching procedure using, in each step, the value of CIT paid in addition to the firm characteristics used in rounds 1-3; adding this new matching variable results in finding single twins in many new cases. As the result, 88.5% of all CIT payers (and 93% of observations in the CIT database) for 2008–14 are matched. 4.3. Quality of Matching For the whole period, 2008–14, matched firms account for 93% of all observations in the CIT data; coverage ranges from 90% in 2008 to 94% in 2012–13 (Table 4). Descriptive statistics (see Annex 3) suggest that matched firms are representative of all firms. Indeed, distributions of the two sets by 2-digit NACE sectors, by region, and by type of settlement are very similar, as are also distributions by profit before taxes (see Tables A3.1 – A3.4 for details). Table 4. Firms with Matched CIT and Annual Report Data, 2008–14 Year Total Matched Matched; Fixed Assets not Missing # obs # obs % of CIT # obs % of % of Firms Declaring Some Payers CIT Asset Depreciation Payers 2008 76,578 68,776 89.8 55,496 72.5 90.2 2009 77,126 70,792 91.8 55,052 71.4 90.9 2010 80,306 74,787 93.1 55,737 69.4 91.1 2011 82,021 76,787 93.6 54,521 66.5 91.0 2012 84,396 79,474 94.2 54,374 64.4 91.1 2013 86,895 81,853 94.2 54,443 62.7 91.6 2014 87,822 82,336 93.8 54,194 61.7 92.2 Total 575,144 534,805 93.0 383,817 66.7 91.1 Source: Calculation with MOF firm-level CIT data and annual report data provided by Lursoft IT. However, matching CIT and annual report data is not our main purpose: what we need are data on tangible assets and derived investment data. Unfortunately, data on tangible assets are missing for about 25% of matched firms in 2008–09, 20% in 2010, and 33% in 2011–14. As a result, the working sample (matched firms with non-missing data on tangible assets) covers about 9 In the third round, new matches emerge because some firms from CIT data that had more than one match in the annual reports data during the second round are left with only one twin after completing the second round. 10 55,000 firms in each of the study years, from 72% of all firms in the CIT dataset in 2008–09 to 62% in 2014 (Table 4). The working sample coverage is much higher (above 90%, see Table 4, last column) for those firms in the CIT dataset whose balance sheets declare some asset depreciation (plausibly, most firms with non-negligible fixed assets do so). This suggests that for purposes of this paper the working sample is representative. The set of firms declaring some asset depreciation does not differ significantly from the whole CIT dataset in terms of distribution by 2-digit NACE code (see Table A3.1, right panel), region, and type of settlement. The same is true for the working sample (results available on request). This suggests that there are no major objections to the idea that the working sample is representative of all firms with non-negligible fixed assets. On the other hand, the profits before taxes of firms in the working sample (and of those declaring some asset depreciation) are on average somewhat larger and more widely dispersed than those of all firms in the CIT dataset or all matched firms, see Table A3.4. 5. Econometric Methodology 5.1.The Difference-in-Differences Approach To evaluate the effect of a firm’s past AD experience on its investment we estimate fixed-effects panel data models of the following type: Yit = t + βtZ_ADit-1 + γXit-1 + ui + εit (1) where Yit is a measure of investment by firm i in year t, t are time fixed effects, Z_ADit-1 is a measure of AD used by firm i in the previous year, βt are time-varying effects of AD on investment, Xit-1 is a lagged vector of firm characteristics (including fixed assets, employment, turnover, profit, and firm age), ui are unobserved firm fixed effects, and εit are error terms. To simplify notation, we allow t in (1) to take all values including the one for the reference year, say, 1. In model (1), the AD variable can suffer from endogeneity caused by reverse causality (a firm planning to invest in t may want to make use of AD in t − 1) or by time-varying unobserved factors affecting both investment and AD decisions. Our baseline models use investment and AD rates rather than just indicators of positive investment and positive AD value, which arguably makes the endogeneity risk less significant. However, to address the endogeneity problem we proceed as follows: 1. Construct the treatment group T, firms whose investment behavior is likely affected by the AD policy, and the control group C, firms probably not significantly affected by the policy (see details below). 2. Estimate (1) separately on T and C. 3. Apply the difference-in-differences methodology by comparing the change in βt (vs. the base year) in the treatment group with the corresponding change in the control group. Technically, this is equivalent to estimating on the pooled (T and C) sample a 11 fixed-effect model like (1) amended with the treatment group dummy (also denoted T ) and its interactions with other variables: Yit = t + βtZ_ADit-1 + tT+ tT×Z_ADit-1 + γXit-1 + T×Xit-1 + εi (2) The coefficients of interest in (2) are t; these are equal to differences in βt from (1) estimated on treatment and control groups. As argued next, if the base year corresponds to the crisis period, significant and positive t in other periods will indicate that AD has a positive effect on firm investment. We then modify this version of regression discontinuity design (RDD) by replacing the variation across time with the variation (within treatment and control groups) across firm size (in terms of employment). Model (2) then is replaced by a model with size-specific effects Yist = st + βsZ_ADit-1 + sT + sT×Z_ADit-1 + γsX_it-1 + sT×X_it-1 + ui + εit , (3) where s varies across size categories and X_ includes firm characteristics other than employment. If one can argue that in the base category the treatment effect is absent, significant and positive s will indicate that AD has a positive effect for firms in other categories. Finally, we estimate a model similar to (3) where s varies by firm category in volatility of turnover, and X_ includes other firm characteristics, size among them. 12 5.2. Identification Strategy We follow Zwick and Mahon (2017) and use as the control group firms in industries where investment is mostly short-term as the control group, as AD only modestly alters their depreciation schedule.10 Technically, our analysis differs from that of Zwick and Mahon (2017), which assigns an industry to the treatment group if in the industry the average discounted value of one dollar of investment deductions without bonus depreciation is low, and to the control group if it is high). We use for the same purpose the industry average11 ratio DR of accelerated depreciation value to balance sheet depreciation value, and the industry average difference DD of the same quantities. Our treatment group features high values of DR and DD and our control group has low values (see Table A4.1 in Annex 4 for details). When all types of AD are considered together, the treatment group includes firms belonging to the top 20% in terms of DR (i.e., having DR ≥ 1.275) and to the top 33% in terms of DD (i.e., having DD ≥ €3,000), which ensures that AD provides a non-negligible increase in tax deductions. On the other hand, the control group includes firms from industries with either DR < 1.05 or DD < €500 and hence the increase in tax deductions due to AD is small. Our estimates of the AD effect on investment include firm fixed effects and also control for a number of time-varying firm characteristics. Nevertheless, in the spirit of the D-i-D methodology, it is preferably that the control and the treatment groups be as similar as possible. Tables A4.2– A4.3 show that distributions of treatment and control groups by firm size, turnover, investment rate, region and type of settlement are quite similar. Moreover, across the years the average proportion of firms making an investment of at least €100 is about 50% in both groups, and average investment rate (with respect to beginning-of-the-year capital) is 24.3% in the treatment group and 27.9% in the control group.12 Because AD began before our sample period, we are not able to use its introduction as a natural experiment, in which case t = 1 in (1) would correspond to the pre-reform period.13 Instead, we argue that during the crisis in 2009–10 investment rates were very low (see Figures A1–A2) due to cash constraints and the uncertain prospects of the economy, and hence the effect of past AD experience on investment was absent or very small in both groups (see Figure 3 for empirical evidence). Hence, this period can be used as a quasi-counterfactual. Post-crisis, investment activity in Latvia revived, but not to the pre-crisis level (Figures A1–A2); firms are more often cash constrained and are more careful in making decisions to invest. We expect that in comparison 10 We thank Irem Guceri for this idea. 11 Like Zwick and Mahon (2017), we use four-digit industries. 12 With treatment and control groups defined by the industry they belong to, differences in this respect are inevitable. For example, real estate activities account for more than half of the treatment group observations, while retail trade, and trade and repair of motor vehicles, account for one-third of the control group. Most other services are almost completely in the control group, although manufacturing, utilities, and construction firms are found in both groups. 13 As mentioned in Section 2, the AD component for SAAs experienced eligibility changes in 2010–11, but several factors make it complicated to use this as a natural experiment. Among them, first, the 2010 reform has made it much easier to apply AD, but in 2011 many eligible municipalities were removed from the list. Second, firms that lost eligibility could continue to use AD for items purchased while they were still eligible. 13 with 2009, the increase of the effect of past AD experience on investment is much larger in the treatment than in the control group, so the post-crisis t in model (2) are positive. Figure 3 supports this hypothesis. A weakness in this approach is that we cannot convincingly test the parallel lines assumption because we only have 2 crisis years. Our second version of RDD refers to variation across firm size instead of across time. Small firms are more likely to be cash-constrained and to have uncertain prospects, hence we expect that in the treatment group, the effect of past AD experience on investment increases as firm size (measured by employment) falls. By contrast, in the control group, because small firms do not see an immediate cash benefit from AD, they are likely to invest only sporadically if at all. For large firms in the control group investment is likely to be a part of their business model; plausibly, these investments are regular due to short asset lives, leading to reversed causality of the estimated AD effect, which is likely to increase with firm size (larger firms invest more regularly). Using large (50+ worker) firms as the base category, we estimate model (3) and indeed find a positive effect of AD on investment for smaller firms in the treatment group (Figure 4). In the treatment group, the effect gets larger as firm size falls, suggesting that there is no reversed causality. In the control group, the estimated effect grows with firm size, suggesting that reversed causality is at work. The D-i-D effect is highly significant for firms with fewer than six workers. Our third version of RDD refers to the uncertainty-mitigating effect of accelerated depreciation and uses variation across groups with low, medium, and high volatility of turnover. We expect that in the treatment group, the effect of past AD experience on investment increases with volatility of turnover, while in the control group the uncertainty factor is unlikely to have a significant impact on a priori weak AD effect. The results presented in Figure 5 below support this expectation; the D-i-D effect is highly significant for firms with medium and high volatility of turnover (and larger among firms with high volatility). 6. Results 6.1 Key variables We define a firm’s investment rate in year t as Inv_rate (t) = log(1 + (Gross investment in fixed assets)(t))/ (K(t – 1)) (4) where K(t) is the end-of-year value of fixed assets, and (Gross investment in fixed assets)(t) = K(t) – K(t–1) + Balance sheet depreciation value (t). (5) Note that (4) is just the continuously compounded version of the usual investment rate. We have used two firm-level measures of AD experience: the accelerated depreciation rate: AD_rate (t) = log(1 + ADV(t)/K(t – 1)), (6) and the reduction in taxable income caused by AD (also scaled by K(t–1)): AD_gain (t) = log (1+ (ADV(t) – BDV(t))/K(t – 1)). (7) The rationale for using AD_gain is straightforward; use of the AD_rate is motivated by behavioral considerations: the tax declaration template refers to ADV as one of the items reducing taxable 14 income, so managers or owners of small firms might see this (rather than AD_gain) as a measure of the tax benefit. In models with Inv_rate as the dependent variable, estimated coefficients on lagged AD_rate or AD_gain can be interpreted as investment elasticities with respect to the previous year AD value or the corresponding reduction in taxable income. Both AD_rate and AD_gain can be decomposed into three components corresponding to AD of new equipment, AD in SAAs, and other types of AD. Accordingly, every model can be estimated in four specifications: with total AD_rate, decomposed AD_rate, total AD_gain, and decomposed AD_gain. 6.2 Total aggregate effects of accelerated depreciation Estimation results from specifications (1) and (2) can be found in Table A5.1; Figure 3 displays the corresponding effects of AD. Figure 3. Estimated Effect of Past AD Experience on Firm Investment. Treatment and Control Groups, D-i-D: (T_year – T_2009) – (C_year – C_2009) A. Effect of Lagged AD Value. 15 B. Effect of Lagged AD-caused Reduction in Taxable Income Notes: Labels in the D-i-D series show t-values from fixed-effect model with interactions. As expected, in both treatment and control groups and for both AD value and tax gain, lagged AD variables have no effect on firm investment in 2009–10. In 2011–14, we find a positive and significant effect for the treatment group. In the control group, there is no effect of the past AD value; the past AD gain appears to be significant in 2012 –14, but the effect is much weaker than in the treatment group. Both specifications produce a strongly significant positive D-i-D effect in 2011–14. As far as other factors are concerned, the investment rate tends to increase with firm size14 and turnover, other things being equal. Lagged capital stock has a negative effect on investment. Not surprisingly, enterprises in the first two to three years after registration invest the most. Among profit-making enterprises, as might be expected, investment rises in track with the previous year’s profit—but loss-making firms also tend to invest more the larger the previous year’s loss15. It is not unreasonable for firms with large losses to invest more than those with small losses, but that this might also relate to losses carried forward and coordination of investment plans with a tax-optimization strategy. 6.3 Effects of accelerated depreciation by firm size and by volatility of turnover Figure 4 presents the estimation results for specification (3) with identification based on variation across firm size groups. 14 However, zero-worker firms are more likely to invest than firms with 1–5 workers. 15 In the treatment group, this effect is somewhat smaller than the similar effect among profit-makers; in the control group the effect among loss-makers is much weaker than among profit-makers. 16 Figure 4. Estimated Effect of Past AD Experience on Firm Investment, by firm size. Treatment and Control Groups, D-i-D: (T_size – T_50+) – (C_size – C_50+) A. Effect of Lagged AD Value B. Effect of Lagged AD-caused Reduction in Taxable Income Notes: Labels in the D-i-D series show t-values from fixed-effect model with interactions. In the treatment group, there is a positive and significant effect of lagged AD value or lagged AD- caused reduction in taxable income on investment by firms with fewer than 50 workers, except for firms with 6–10 workers. This effect gets larger as firm size falls, suggesting that there is no reversed causality. In the control group, the estimated effect of AD-caused reduction in taxable income is positive and significant among firms with at least one worker; the effect of AD_rate is positive and significant for firms with more than 10 workers, while among firms with 6–10 workers it is almost significant. This effect grows with firm size, suggesting that reversed causality is at work (firms use AD because they invest regularly) in the control group. In both specifications, 17 the D-i-D estimate of AD effect on investment rate is highly significant for firms with fewer than six workers. Figure 5 presents the results of specification (3) with identification based on variation across turnover volatility groups. Figure 5. Estimates Effects of Past AD Experience on Firm Investment, by turnover volatility. Treatment and Control Groups, D-i-D: (T_high – T_low) – (C_high – C_low) A. Effect of Lagged AD Value. B. Effect of Lagged AD-caused Reduction in Taxable Income Notes: Labels in the D-i-D series show t-values from fixed-effect model with interactions. Among firms with low turnover volatility, there is no AD effect; firms with medium and high volatility show a positive and significant effect of past AD experience (AD_rate or AD-caused 18 reduction of taxable income) on investment. The latter effect rises with turnover volatility, and the estimated D-i-D effect is highly significant. 6.4 Disaggregated AD effects on firm investment rate and investment decisions, by program type Table A5.2 presents the main results from versions of models (1) and (2) with three separate variables for new equipment (AD rate_new), SAAs (AD_rate_terr), and other types of AD (AD_rate_oth) instead of the total AD_rate. For the latter (referred to as “the general AD scheme” in Section 2 above), as well as for the AD of new equipment, we find, like for the total AD_rate, positive effects on investment rate in the treatment group during the post-crisis period 2011- 2014. However, corresponding D-i-D estimates are significant only for the general scheme (coefficients on AD_rate_oth are close to the ones reported in Table A5.1 for the total AD_rate and even more significant). The D-i-D estimate of AD_rate_new is significant (and positive) only in 2014 (although in the treatment group significant effects are found in 2013-2014), and the D-i-D estimates of AD_rate_terr are not significant at all.16 This is likely because only small share of firms were engaged in AD of new equipment and AD in SAAs (see Table 2 and Table A2.2)17, so the control and treatment groups should be redefined to isolate these effects (if any). We come back to this later in this section. Using the same sample of treatment and control groups, we also estimated linear probability models (1) and (2), where dependent variable Yit is an indicator that firm i invests at least €100 in year t. Table A5.3 presents the results from models with disaggregated AD effects (results for the total AD effect are briefly discussed below). We find that in treatment and control group alike, both the total AD_rate and AD_rate related to the general AD scheme (AD_rate_oth) have a positive effect on probability to invest next year, but the D-i-D estimates are never significant, so the effect cannot be claimed causal. In the treatment group, moreover, AD_rate_terr has a positive and significant effect on investment decisions in 2010, when the restrictive project requirement for this category of AD was lifted. This supports the descriptive evidence from Table 2. However, here the D-i-D estimate is again not significant. Estimates presented in Table A5.3 do not suggest there is any effect of the rate of AD of new equipment (AD_rate_new) on probability to invest next year, but this, plausibly, is because the “total” Treatment and Control groups are not suitable for this (see below). Effects of other variables18 on probability to invest have the same signs as the effects on investment rate reported in Table A5.1 (an exception is lagged loss, which is not significant). Replacing AD_rate with AD_gain variables, as well as changing the investment threshold from 100 euro to 1000 euro does not change the conclusions. 16 Replacing AD_rate with AD_gain does not change the situation. 17 Moreover, for AD in specially assisted areas the eligibility is based on territorial units, and the number of eligible firms in the control group for the total AD (see Table A4.1), which is used in estimates reported in Table A5.2, is substantially larger than in the treatment group. In this regard, the estimated positive effects of AD_rate_terr reported in Table A5.2 in the control group in 2011, 2012 and 2014 is noteworthy. 18 To save space, Table A5.3 omits these effects. 19 To isolate the effect of the AD of new equipment, we narrow down the Treatment group used until now (T_tot) by imposing, in addition, the following conditions: (i) the industry19 average accelerated depreciation value of new equipment ADV_new ≥ 1000 euro; (ii) the industry average ratio of the accelerated depreciation value to the balance sheet depreciation value of new equipment DR_new ≥ 2.00; (iii) there are more than three observations with positive ADV_new in the industry. Under these conditions, one can expect substantial tax savings from using the AD of new equipment for firms in the restricted Treatment group, T_new. Together, these conditions reduce the number of firms in T_tot by nearly a half, from 10.6 to 6.2 thousand (Table A4.1). The restriction on the total Control group (C_tot) that the industry average difference DD = ADV − BDV ≤ 500 EUR applies of course also to DD_new and is sufficient to make non-negligible tax savings unlikely. To ensure sufficient common support with the Treatment group T_new, we require, in addition, the industry average share of firms using AD of new equipment in C_new to be at least 0.38% (the minimum in T_new). This reduces the number of firms by one-third, from 30.9 in C_tot to 20.5 thousand in C_new (Table A4.1). Tables A4.2-A4.3 show that T_new and C_new do not differ much in terms of distribution by firm size, turnover, investment rate, region and type of settlement. Disaggregated by type of AD specifications (2) and (3) estimated on pooled T_new and C_new samples provide some (inconclusive) evidence of the effect of the past AD of new equipment on firm’s investment rate and investment decisions. In (2), with identification across periods, the D-i-D estimate of AD_rate_new is significant (at 10%) only for 2014, while AD_rate_oth is significant (at 1%) for 2011-2014 (these results are not presented to save space). Table A5.4 presents results from model (3), with identification across size groups. AD of new equipment is found to have a significant (at 5%) effect both on next year investment rate and on probability to invest next year, but only for firms with six to ten workers; the effect on investment rate is also positive and close to being significant for firms with 11 to 49 workers. Finally, to isolate the effect of AD in specially assisted areas (SAAs), we restrict the “total” Treatment and Control groups to the subsample of firms which during some years between 2007 and 2010 operated in territories which had the SAA status but has lost it either in 2010 or (in most cases) in 201120 (see Figure 1). This choice is driven partly by data limitations: while we can identify all cases when firms used the SAA component of the AD policy, we cannot identify all eligible firms because in about 14% of our working sample the territory codes are not detailed enough. This limitation does not apply to the category of SAAs used for the identification. Furthermore, based on the results presented in Figure 4, we restrict the analysis by firms with 1 to 5 workers (as of the previous year); see Table A4.1 for formal definition of the Treatment and Control groups (T_SAA and C_SAA). Tables A4.2-A4.3 confirm that T_SAA and C_SAA are quite similar in terms of distribution by region, type of settlement, turnover and investment rate. Fixed effects panel data linear probability models in form (1) and (2) find significant (at 5%) positive D-i-D effects of the rate of AD in SAAs in the previous year on probability of investing at 19 As before, we use four-digit level industry. 20 These firms could continue to use AD for items purchased under the SAA status. 20 least 100 EUR (or at least 1000 EUR) in 2010 and 201221. This suggests that after lifting the restrictive project requirement for AD in SAAs in 2010, this component of the AD policy had a positive effect on investment decisions for small firms. 7. Conclusion This paper has studied the effect of generous accelerated depreciation (AD) policy on firm investment in Latvia in 2009–14. Lacking data for a natural experiment, we use difference-in- differences methodology with identification based on variation across time, with the crisis period serving as quasi-counterfactual; by firm size, with large firms unlikely to be genuinely affected by the policy; and by turnover volatility, with low-volatility firms unlikely to be affected. We find a positive effect of the past use of the general AD scheme on firms’ investment rate, and the effect is stronger in industries with most of their investment in long-lasting assets, in enterprises with fewer than six employees, and in firms with high turnover volatility. Regarding AD of new equipment, we find a significant effect both on next year investment rate and on probability to invest next year, but only for firms with six to ten workers. After lifting the restrictive project requirement for AD in specially assisted areas, this category of AD has boosted investment decisions of firms with 1 to 5 workers. Our results indicate that AD policy has indeed stimulated investment in Latvia. However, the evidence for the effect is stronger for the general scheme than for new equipment and for specially assisted areas, and this raises the question whether the AD scheme in Latvia was too general and/or too generous, resulting in over-investment in less productive firms or industries. 21 The effect of the general AD scheme remains positive as well, but AD of new equipment has no effect. These results are available on request. 21 References Bronzini R., G. de Blasio , G. Pellegrini, and A. Scognamiglio. 2008. “The Effect of Investment Tax Credit: Evidence from an Atypical Programme in Italy.” Banca D’Italia WP N661, Rome. https://www.bancaditalia.it/pubblicazioni/temi-discussione/2008/2008- 0661/en_tema_661.pdf?language_id=1. Desai, M. A., and A. D. Goolsbee. 2004.”Investment, Overhang, and Tax Policy.” Brookings Papers on Economic Activity 2: 285–338. Devereux. M..P., G. Maffini, and J. Xing. 2016. “The Impact of Investment Incentives: Evidence from UK Corporation Tax Returns. Oxford University Centre for Business Taxation Working Paper Series. WP16/01. Oxford, UK. Edgerton J. 2010. “Investment Incentives and Corporate Tax Asymmetries.” Journal of Public Economics 94 (11-12): 936–52. House, C. L., and M. D. Shapiro. 2008. “Temporary Investment Tax Incentives: Theory with Evidence from Bonus Depreciation.” American Economic Review 98 (3): 737–68. http://www.aeaweb.org/articles.php?doi=10.1257/aer.98.3.737. Jacobs, B., E. Sinnott, E. Skrok, M. Hazans et al. 2017. “Latvia Tax Review”. Washington, D.C. : World Bank Group. http://documents.worldbank.org/curated/en/587291508511990249/Latvia-tax-review Jorgenson, D. 1963. “Capital Theory and Investment Behavior.” American Economic Review 53 (2): 247– 59. Kaplan S. N., and L. Zingales. 1997. “Do Investment-Cash Flow Sensitivities Provide a Useful Measure of Financing Constraints?” Quarterly Journal of Economics 112 (1):169–215. Mills, L. F., K. J. Newberry, and W. B. Trautman. 2002. “Trends in Book-Tax Income and Balance Sheet Differences.” Tax Notes 96 (8): 1109–25. Mosberger, P., and Z. Varga. 2018. Aid intensity and Choice of Location for Subsidiaries Investment in Hungary. Manuscript, World Bank, Washington, DC. Ohrn, E. (forthcoming). “The Effect of Tax Incentives on U.S. Manufacturing: Evidence from State Accelerated Depreciation Policies,” Journal of Public Economics. Ohrn, E. 2018. The Effect of Corporate Taxation on Investment and Financial Policy: Evidence from the DPAD. American Economic Journal: Economic Policy 10 (2): 272 - 301 Skrok, E., et al. 2018. ”Investment Tax Incentives in Poland: A Quasi-Experimental Evaluation.” Manuscript, World Bank, Washington, DC. Stein, J. C. 2003. “Agency, Information and Corporate Investment.” In Handbook of the Economics of Finance, Vol. 1, edited by G. M. Constantinides, M. Harris, and R. M. Stulz, pp. 111–65. Amsterdam: Elsevier. Yagan, D. 2015. “Capital Tax Reform and the Real Economy: The Effects of the 2003 Dividend Tax Cut. “American Economic Review 105 (12): 3531–63. World Bank (2018). Zwick E., and J. Mahon. 2017. “Tax Policy and Heterogeneous Investment Behaviour,” American Economic Review 107(1): 217–48. https://doi.org/10.1257/aer.20140855. 22 Annex 1. The Economic Context: Firm Investment in the Baltic Countries and Poland In 2009–15, the business investment rate (gross investment as percentage of the gross value-added of nonfinancial corporations) in Latvia was largely in line with other Baltic countries and Poland, although before the crisis (in 2008) the investment intensity in Latvia was well above its level in neighboring countries (Figure A1, A). However, since 2010, Latvia has trailed Estonia in terms of investment per person employed, and the gap is widening (FigureA1, B). This is true for most sectors (Figure A2). Figure A1. Investment intensity in Total Business Economy. The Baltic countries and Poland, 2008-15. A. Business Investment Rate, Percent B. Investment per Person Employed, €thousands A B 60 10 8 40 6 4 20 2 0 0 2008 2009 2010 2011 2012 2013 2014 2015 2008 2009 2010 2011 2012 2013 2014 2015 LV LT EE PL LV LT EE PL Sources: Calculation with Eurostat data. Note: Total business economy includes all NACE activities except sections A, K, O, P, Q, R, S94, S96, T, U. Country notations: LV – Latvia, LT – Lithuania, EE – Estonia, PL – Poland. Since 2008, gross investment in tangible goods in the Latvian business economy22 plunged, hitting bottom in 2010. Despite some recovery thereafter, by 2015 the investment intensity was less than a half of its level in 2008 (Figure A1, A and B). Up to 2012, business investment rate was higher in Latvia than in other Baltic countries and Poland. Since 2013, investment in the Baltics states has been getting closer, but in Poland is still lower than in any of the Baltic countries see Figure A1, A). In 2009–15, investment per person employed in business economy in Estonia was well above that in Latvia, Lithuania, and Poland. As of 2009, Latvia lost its leadership position, and in 2015 investment there fell slightly below investment per person employed in Lithuania and Poland (Figure A1, B). Investment per Person Employed by Industry In Latvia, the most investment resources per person employed are found in utilities (NACE divisions s D and E), as well as real estate activities (NACE division L). In most industries, investment per person is very close in Latvia, Lithuania, and Poland, but Estonia significantly outstrips its neighbors in almost all sectors (Figure A2). 22 NACE: B-N_S95_X_K: Total business economy; repair of computers, personal and household goods; except financial and insurance activities. This aggregate includes all NACE activities except sections A, K, O, P, Q, R, S94, S96, T and U 23 Figure A2. Investment per Person Employed, Selected Industries, the Baltics and Poland, € thousand 8 Manufacturing (C) 150 Electricity, gas, steam and air conditioning 6 supply (D) 100 4 50 2 0 0 2008 2009 2010 2011 2012 2013 2014 2015 2008 2009 2010 2011 2012 2013 2014 2015 LV LT EE PL LV LT EE PL 60 Water supply; sewerage, waste 4 Construction (F) management and remediation activities (E) 3 40 2 20 1 0 0 2008 2009 2010 2011 2012 2013 2014 2015 2012 2013 2014 2015 LV LT EE PL LV LT EE PL 5 Wholesale and retail trade; repair of motor 15 Information and communication (J) 4 vehicles and motorcycles (G) 10 3 2 5 1 0 0 2008 2009 2010 2011 2012 2013 2014 2015 2008 2009 2010 2011 2012 2013 2014 2015 LV LT EE PL LV LT EE PL 80 Real estate activities (L) 8 Professional, scientific and technical 60 6 activities (M) 40 4 20 2 0 0 2008 2009 2010 2011 2012 2013 2014 2015 2008 2009 2010 2011 2012 2013 2014 2015 LV LT EE PL LV LT EE PL Sources: Eurostat data. Investment Rate by Industry Investment rate in Latvia was higher than in the other Baltic countries and Poland in manufacturing (up to 2013), construction (2012 and 2014), and real estate activities (up to 2012). 24 Figure A3. Investment Rate in Selected Industries, the Baltics and Poland, Percent 60 Manufacturing (C) 150 Electricity, gas, steam and air conditioning supply (D) 40 100 20 50 0 0 2008 2009 2010 2011 2012 2013 2014 2015 2008 2009 2010 2011 2012 2013 2014 2015 LV LT EE PL LV LT EE PL 200 Water supply; sewerage, waste 25 Construction (F) 150 management and remediation activities (E) 20 100 15 50 0 10 2008 2009 2010 2011 2012 2013 2014 2015 2012 2013 2014 2015 LV LT EE PL LV LT EE PL 30 Wholesale and retail trade; repair of motor 30 Information and communication (J) 25 vehicles and motorcycles (G) 25 20 20 15 15 10 10 2008 2009 2010 2011 2012 2013 2014 2015 2008 2009 2010 2011 2012 2013 2014 2015 LV LT EE PL LV LT EE PL 250 Real estate activities (L) 40 Professional, scientific and technical 200 30 activities (M) 150 20 100 50 10 0 0 2008 2009 2010 2011 2012 2013 2014 2015 2008 2009 2010 2011 2012 2013 2014 2015 LV LT EE PL LV LT EE PL Sources: Eurostat data. Figure A4 features relationship between investment per person employed and apparent labor productivity (measured as gross value added per person employed) in manufacturing in Latvia and Poland in 2008-15, separately for high technology, medium-high technology, medium-low technology, and low-technology industries. Figure A5 shows the relationship between business investment rate and apparent labor productivity (gross value added per person employed) for the same groups of industries in Latvia and Poland, in 2008–15. 25 The relationship between investment per person employed and apparent labor productivity is stronger in Poland than in Latvia (Figure A4), though for both the relationship between investment and apparent labor productivity is weak (Figure A5). Figure A4. Investment per Worker and Apparent Labor Productivity by Level of Technology in Manufacturing, Latvia and Poland, 2008–15, € Thousand Latvia Poland 12 9 8 10 7 Investment 8 Investment 6 per person per person 5 6 employed, employed, 4 mln. euro 4 mln. euro 3 2 2 1 0 0 10 20 30 40 50 10 20 30 40 50 Apparent labour productivity, thousand euro Apparent labour productivity, thousand euro High-technology m. Medium high-technology m. High-technology m. Medium high-technology m. Medium low-technology m. Low-technology m. Medium low-technology m. Low-technology m. Surces: Eurostat data. Note: (1) High-technology manufacturing (C_HTC) covers NACE activities:C21, C26 and C30.3; (2) medium-high (C_HTC_M) covers NACE activities C20, C25.4, C27 to C29, C30 (except C30.1 and C30.3) and C32.5; (3) medium-low (C_LTC_M) covers NACE activities C18.2, C19, C22 to C24, C25 (except C25.4), C30.1 and C33;and (4) low-technology manufacturing (C_LTC) covers NACE activities C10 to C17, C18 (except C18.2), C31 and C32 except C32.5). . 26 Figure A5. Business Investment and Apparent Labor Productivity by Level of Technology in Manufacturing, Latvia and Poland, 2008–15 Latvia Poland 80 35 70 30 60 25 50 Investment Investment 20 40 rate, % rate, % 15 30 20 10 10 5 0 0 10 20 30 40 50 10 20 30 40 50 Apparent labour productivity, thousand euro Apparent labour productivity, thousand euro High-technology m. Medium high-technology m. High-technology m. Medium high-technology m. Medium low-technology m. Low-technology m. Medium low-technology m. Low-technology m. Source: Eurostat data. Notes: The business investment rate equals gross investment divided by gross value added of non-financial corporations. Data for Poland for 2008 and 2012 are missing. (1) High-technology manufacturing (C_HTC) covers the following NACE activities: C21, C26 and C30.3. (2) Medium-high (C_HTC_M) covers NACE activities C20, C25.4, C27 to C29, C30 (except C30.1 and C30.3) and C32.5. (3) Medium-low (C_LTC_M) covers NACE activities C18.2, C19, C22 to C24, C25 (except C25.4), C30.1 and C33. (4) Low- technology g (C_LTC) covers NACE activities C10 to C17, C18 (except C18.2), C31 and C32 (except C32.5). 27 Annex 2. Descriptive Evidence on Accelerated Depreciation in Latvia Table A2.1 Regular and Bonus Depreciation Schedules for 1 million Investment in Five-Year Items: U.S ($). vs Latvia (Euro) Year 0 1 2 3 4 5 Total US Normal depreciation Deductions (000s) 200.0 320.0 192.0 115.2 115.2 57.6 1000.0 Tax benefit (t=35%) 70.0 112.0 67.2 40.3 40.3 20.2 350.0 NPV 70.0 104.7 58.7 32.9 30.8 14.4 311.4 Bonus depreciation (50 percent) Deductions (000s) 600.0 160.0 96.0 57.6 57.6 28.8 1000.0 Tax benefit (t=35%) 210.0 56.0 33.6 20.2 20.2 10.1 350.0 NPV 210.0 52.3 29.3 16.5 15.4 7.2 330.7 Present value payoff 19.3 Latvia Balance sheet depreciation (Double declining balance method or 2x straight line) Deductions (000s) 200.0 320.0 192.0 115.2 69.1 103.7 1000.0 Tax benefit (t=35%) 70.0 112.0 67.2 40.3 24.2 36.3 350.0 NPV 70.0 104.7 58.7 32.9 18.5 25.9 310.6 Depreciation for taxation purpose (35% rate, doubled) Coefficient of 1.5 is applied to asset value before depreciation for tax purposes Deductions (000s) 525.0 682.5 204.8 61.4 18.4 5.5 1497.6 Tax benefit (assuming t=35%) 183.8 238.9 71.7 21.5 6.4 1.9 524.2 NPV 183.8 223.2 62.6 17.5 4.9 1.4 493.4 Present value payoff (if t=35%) 182.8 Present value payoff (actual t=15%) 78.4 28 Table A2.2. Reduction in Taxable Corporate Income due to Accelerated Depreciation of New Equipment, €1,000 Top 25 Sectors, Annual Average, 2012–14 NACE Sector Code Total Mean p50 p75 p90 # firms Manufacturing 10-33 88,040 184 7 75 35 478 Wood products 16 31,899 288 7 66 548 111 Food products 10 15,254 218 20 186 666 70 Basic Metals 24 12,920 3,524 161 376 3,097 4 Metal Products 25 5,074 89 7 84 231 57 Printing & reproduction 18 4,270 200 6 26 474 21 Rubber & plastic products 22 3,207 182 19 135 460 18 Other nonmetallic products 23 2,615 76 0 45 216 34 Pharmaceutical 21 2,300 431 6 308 1,802 5 Beverages 11 2,251 193 61 320 567 12 Electronic and optical products 26 1,350 109 4 122 229 12 Chemicals & chemical products 20 1,247 87 8 39 181 14 Textiles 13 994 107 2 44 271 9 Machinery & equipment n.e.c. 28 887 67 18 113 255 13 Electrical equipment 27 864 216 23 360 852 4 Furniture 31 753 27 5 33 84 28 Other sectors Electricity, gas, steam 35 86,356 1364 120 371 1,048 63 Real estate 68 8,352 234 1 7 150 36 Agriculture 1 6,316 95 11 68 191 66 Waste collection 38 2,587 268 33 219 1,045 10 Forestry 2 2,303 93 4 96 210 25 Land transport 49 2,113 60 2 24 77 35 Water collection & supply 36 1,686 632 347 1,002 2,210 3 Rental & leasing activities 77 1,310 179 8 57 156 7 Other mining and quarrying 8 1,114 68 10 60 203 16 Specialized construction 43 868 26 1 9 27 34 Sources: Ministry of Finance firm-level CIT data. 29 Annex 3. Matched Firms and All Firms: Distribution of Key Characteristics Table A3.1 2-digit NACE Sector Share Differences: Distribution Parameters, 2008–14 Percentage Points Matched firms vs all firms declaring some Matched firms vs all firms in CIT data asset depreciation 2008 2009-2010 2011-2012 2013-2014 2008 2009-2010 2011-2012 2013-2014 p50 0.01 0.01 0.01 0.02 0.01 0.02 0.03 0.03 p75 0.02 0.02 0.04 0.05 0.05 0.05 0.09 0.11 p90 0.05 0.06 0.12 0.10 0.18 0.21 0.24 0.33 p95 0.08 0.10 0.18 0.15 0.27 0.37 0.48 0.46 max 0.14 0.27 0.76 0.49 1.22 1.23 2.09 1.79 mean 0.02 0.02 0.05 0.05 0.07 0.08 0.12 0.13 sd 0.03 0.04 0.11 0.08 0.15 0.18 0.28 0.28 N 95 95 95 95 95 95 95 95 Source: Firm CIT and annual report data. Table A3.2 Distribution of Firms by Region, 2008–14 Percentage Points Full CIT database Matched firms 2008 2009-2010 2011-2012 2013-2014 2008 2009-2010 2011-2012 2013-2014 Riga 56.9 56.2 55.2 54.5 56.0 56.0 54.9 54.1 Pieriga 15.5 16.1 16.6 17.1 16.0 16.4 16.8 17.3 Kurzeme 8.0 8.0 8.1 8.1 8.1 8.0 8.2 8.3 Latgale 6.7 6.7 6.7 6.7 6.4 6.4 6.6 6.6 Vidzeme 6.4 6.5 6.6 6.7 6.5 6.6 6.7 6.8 Zemgale 6.4 6.5 6.8 6.8 6.6 6.6 6.8 6.9 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Source: Firm CIT and annual report data. Table A3.3 Distribution of firms by type of settlement, 2008-2014 Percentage points Full CIT database Matched firms 2008 2009-2010 2011-2012 2013-2014 2008 2009-2010 2011-2012 2013-2014 Riga 56.9 56.2 55.2 54.5 56.4 56.0 54.9 54.1 Other main 13.9 13.8 13.5 13.3 13.7 13.6 13.3 13.2 cities Small towns 11.1 11.2 11.1 11.2 11.4 11.4 11.3 11.5 Rural 18.1 18.8 20.3 20.9 18.5 19.0 20.5 21.3 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Source: Firm CIT and annual report data. 30 Table A3.4 2%Trimmed Distribution of Firms by Profit (€1,000) Before Taxes, 2013–14 p10 p25 p50 p75 p90 mean sd N All firms -13.1 -2.1 0.0 5.0 35.0 9.5 60.8 171 223 Matched firms -14.0 -2.4 0.0 5.9 37.1 9.9 61.9 160 907 Matched firms with non-missing fixed assets -18.0 -3.5 0.6 13.2 58.1 16.5 79.5 106 464 All firms declaring some asset depreciation -19.6 -4.0 1.0 15.1 64.5 18.6 89.1 102 222 Source: Firm CIT and annual report data. 31 Annex 4 Treatment and control groups Table A4.1 Definitions of treatment and control groups in terms of 4-digit industry average accelerated depreciation value (ADV) and balance sheet depreciation value (BDV) Accelerated depreciation category Specially assisted areas Total New Equipment (SAAs) Treatment Control Treatment Control Treatment Control ADV/BDV ≥ 1.275 < 1.05 ≥ 1.275 < 1.05 ≥ 1.275 < 1.05 ADV − BDV ≥ € 3000 < € 500 ≥ € 3000 < € 500 ≥ € 3000 < € 500 ADV_new/BDV_new ≥ 2.00 ADV_new ≥ € 1000 N obs with ADV_new > 0 >3 % obs with ADV_new > 0 > 0.38% Firm size (N workers) any any any any 1- 5 1- 5 SAA status no requirements 2007-2010 N firms 10649 30872 6245 20544 430 1092 N obs 59297 158259 36907 106119 1205 3057 Source: Firm CIT and annual report data. Notes: Firm size refers to the previous year. Firms in SAA Treatment and Control groups were operating in territories which had SAA status in 2007-2009 and/or 2010, but lost it since 2010 or (in most cases) 2011; these firms could, however, continue to use AD for items purchased under the SAA status. N firms and N obs refer to the working sample. 32 Table A4.2 Distribution of treatment and control groups by firm size, region and type of settlement Total New equipment SAAs Treatment Control Treatment Control TreatmentControl # workers 0 13.2 8.4 12.2 7.9 1- 5 58.9 60.9 57.8 60.1 100.0 100.0 6 -10 9.9 14.6 10.2 15.3 11 - 49 13.1 13.3 14.9 13.9 50+ 4.8 2.8 4.9 2.8 Total 100.0 100.0 100.0 100.0 100.0 100.0 region Riga 55.3 56.3 52.7 53.7 0.0 0.0 Pieriga 16.5 16.8 16.3 17.0 8.8 6.8 Kurzeme 9.1 7.8 10.0 8.4 19.8 17.8 Latgale 6.1 6.6 6.9 7.4 33.0 36.2 Vidzeme 6.6 6.0 7.2 6.6 28.9 26.9 Zemgale 6.4 6.5 6.9 6.9 9.5 12.3 Total 100.0 100.0 100.0 100.0 100.0 100.0 settlement Riga 55.3 56.3 52.7 53.7 0.0 0.0 Other main cities 13.6 14.5 14.2 14.8 0.0 0.0 Small towns 11.1 12.7 12.0 13.7 31.0 38.9 Rural 20.0 16.5 21.1 17.8 69.0 61.1 Total 100.0 100.0 100.0 100.0 100.0 100.0 Source: Firm CIT and annual report data. 33 Table A4.3 Distribution of treatment and control groups by turnover, investment rate and incidence of investment ≥ €100 Total New equipment SAAs Treatment Control Treatment Control Treatment Control turnover, €1000 p10 4.7 5.6 5.2 6.6 3.3 3.1 p25 17.4 18.7 19.0 22.7 9.8 11.1 p50 77.7 69.8 80.8 79.8 30.4 35.9 p75 337.8 255.7 338.6 276.8 88.4 89.3 mean 78.9 69.4 82.5 78.1 27.9 27.8 investment rate, % p10 -3.3 -2.2 -3.0 -2.4 -1.9 -0.9 p25 0.0 0.0 0.0 0.0 0.0 0.0 p50 0.4 1.3 0.3 1.8 0.0 0.0 p75 21.8 32.0 17.0 28.5 17.9 14.3 mean 24.3 27.9 20.9 24.9 24.6 19.5 Investment ≥ €100, % 50.6 49.6 51.0 50.9 47.3 40.4 Source: Firm CIT and annual report data. 34 Annex 5 Estimation Results – Fixed Effects Panel Data Models Table A5.1 Determinants of Firm Investment Rate with Total AD Effects, 2009–14 Dep. Var.: Inv_rate, AD = AD_rate, see (6) AD = AD_gain, see (7) see (4)-(5) Treatment Control D-i-D Treatment Control D-i-D L.Log(Fixed assets) -0.5017*** -0.3738*** -0.5053*** -0.3843*** 0.0122 0.0054 0.0124 0.0057 L.Log(Turnover) 0.0790*** 0.0767*** 0.0778*** 0.0754*** 0.0083 0.0043 0.0082 0.0043 L.log(Profit) 0.0149*** 0.0108*** 0.0151*** 0.0105*** (if Profit ≥ 1 EUR) 0.0037 0.0019 0.0036 0.0019 L.log(|Loss)) 0.0125*** 0.0033* 0.0124*** 0.0034* (if |Loss|≥ 1 EUR) 0.0036 0.0019 0.0036 0.0019 Firm age (vs. 2-3 yrs) 4-7 -0.0282 -0.0498*** -0.016 -0.0419*** 0.0207 0.0102 0.0205 0.0101 8-10 -0.0677** -0.037** -0.0505* -0.0306* 0.0293 0.016 0.0287 0.0159 11-19 -0.0759** -0.0547*** -0.0482 -0.0485** 0.0383 0.021 0.0375 0.0210 20+ -0.0586 -0.0513* -0.033 -0.0488* 0.0471 0.0264 0.0461 0.0263 L.#employed (vs. 1-5) 0 0.0896*** 0.0398*** 0.0835*** 0.0411*** 0.0234 0.014 0.0232 0.0139 6-10 0.063*** 0.0679*** 0.0590*** 0.0690*** 0.0196 0.0104 0.0195 0.0103 11-49 0.0916*** 0.0905*** 0.0949*** 0.0953*** 0.0282 0.0151 0.0281 0.0149 50+ 0.1230*** 0.1106*** 0.1230*** 0.1179*** 0.0456 0.0277 0.0456 0.0276 L.AD 0.0074 -0.0026 -0.0033 -0.0039 0.0350 0.0140 0.0323 0.0163 year#L.AD: 2010 -0.0369 -0.1109*** 0.0740 -0.0647 -0.0457** -0.019 0.0457 0.0213 0.0504 0.0457 0.0214 0.0505 2011 0.0697 -0.0725*** 0.1422*** 0.0917* -0.0088 0.1005* 0.0457 0.0202 0.0499 0.0474 0.0227 0.0525 2012 0.1494*** 0.0179 0.1315*** 0.1696*** 0.0503** 0.1192** 0.0453 0.0201 0.0496 0.0459 0.0218 0.0508 2013 0.1381*** 0.0184 0.1197** 0.1896*** 0.0782*** 0.1113** 0.044 0.0194 0.0481 0.0462 0.0225 0.0513 2014 0.1267*** 0.0024 0.1243** 0.1705*** 0.0684*** 0.1021* 0.0433 0.0222 0.0486 0.0462 0.0244 0.0522 Year fixed effects yes yes yes yes yes yes Firm fixed effects yes yes yes yes yes yes R-sq: within 0.3252 0.2387 0.2678 0.3167 0.2402 0.2655 overall 0.0753 0.0479 0.0329 0.0679 0.0476 0.034 N obs/N firms, 1000 37.3 / 8.98 103.9 / 26.6 141.2 / 35.6 37.0 / 8.97 103.4 / 26.6 141.4 / 35.5 Source: Calculation with firm CIT and annual report data. Notes: AD refers to total accelerated depreciation variables. Columns “Treatment” and “Control” present estimates of model (1) on groups defined in Table A4.1 (panel “Total”). Columns D-i-D present only time-varying coefficients t of interactions of the treatment dummy (T) with lagged AD_rate or AD_gain from model (2) estimated on the Treatment + Control sample. Robust standard errors clustered on firms in italics. Legend:* p<.1; ** p<.05; *** p<.01. 35 Table A5.2 Disaggregated AD Effects on Firm Investment Rate, by Program Type Dep. Var.: Inv_rate, AD = AD_rate, see (6) AD = AD_gain, see (7) see (4)-(5) Treatment Control D-i-D Treatment Control D-i-D L.AD_new -0.0773 -0.0181 -0.052 0.1007 0.0609 0.1773 0.0622 0.1958 year#L.AD_new: 2010 -0.115 -0.0981 -0.0227 -0.0193 -0.2066 0.1879 0.1431 0.1769 0.2274 0.1735 0.1988 0.2655 2011 0.0314 -0.1797 0.2185 -0.0014 -0.01 0.0158 0.0778 0.2199 0.2333 0.0843 0.2345 0.2489 2012 0.1070 0.0933 0.0221 0.0254 0.0862 -0.0536 0.0735 0.1919 0.2055 0.0818 0.2127 0.2279 2013 0.3441*** 0.2017 0.1506 0.3502*** 0.0742 0.2837 0.1019 0.2301 0.2519 0.1248 0.2045 0.2397 2014 0.4375*** 0.0168 0.4303* 0.4352** -0.0493 0.4911* 0.1449 0.1881 0.2379 0.1748 0.211 0.2743 L.AD_terr -0.2146 -0.3285 -0.3075 -0.0364 0.1785 0.2277 0.2809 0.1252 year#L.AD_terr: 2010 -0.1839 0.1091 -0.2955 -0.1777 -0.0492 -0.1354 0.2289 0.2912 0.3700 0.3119 0.3417 0.4612 2011 0.1386 0.4835* -0.3456 0.4548 0.5557** -0.1034 0.1796 0.2647 0.3193 0.2916 0.2758 0.4003 2012 0.414 0.3919* 0.0183 0.4076 0.0702 0.3361 0.4804 0.2318 0.5329 0.348 0.1395 0.3728 2013 0.2802 0.3672 -0.0886 0.3634 0.0292 0.3330 0.1893 0.2322 0.2988 0.2961 0.1356 0.3233 2014 0.0701 0.4697** -0.4018 0.2478 0.2177 0.0293 0.1941 0.2387 0.3069 0.4771 0.1472 0.4976 L.AD_oth 0.0123 -0.0004 0.0003 -0.0019 0.0359 0.0140 0.0334 0.0162 year#L.AD_oth: 2010 -0.0405 -0.1124*** 0.0716 -0.0678 -0.0468** -0.0209 0.0463 0.0214 0.0510 0.0465 0.0216 0.0513 2011 0.0688 -0.0761*** 0.1450*** 0.0898* -0.0114 0.1013* 0.0468 0.0201 0.0509 0.0500 0.0226 0.0549 2012 0.1462*** 0.0137 0.1330*** 0.1724*** 0.0461** 0.1265** 0.0455 0.0204 0.0499 0.0468 0.0220 0.0517 2013 0.1341*** 0.0141 0.1204** 0.1965*** 0.0751*** 0.1216** 0.0453 0.0194 0.0493 0.0476 0.0226 0.0527 2014 0.1244*** -0.0064 0.1316*** 0.1664*** 0.0606** 0.1063** 0.0445 0.0222 0.0497 0.0477 0.0245 0.0537 Other controls As in Table A5.1 N obs/N firms, 1000 37.2 / 8.98 103.9 / 26.6 141.2 / 35.6 36.9 / 8.96 103.3 / 26.6 140.2 / 35.5 R-sq: within 0.3253 0.2390 0.2681 0.3158 0.2404 0.2653 overall 0.0756 0.0483 0.0327 0.068 0.048 0.0340 Source: Calculation with firm CIT and annual report data. Notes: AD_new, AD_terr and AD_oth refer to AD of new equipment, AD in specially assisted areas and other types of AD, respectively. Otherwise, Notes to Table A5.1 apply. 36 Table A5.3 Disaggregated AD effects on Probability of Investment, by Program Type Dep. Var.: 1 if AD = AD_rate, see (6) AD = AD_gain, see (7) investment ≥ 100 EUR Treatment Control D-i-D Treatment Control D-i-D L.AD_new -0.0083 -0.1286 -0.0194 -0.0473 0.0565 0.1839 0.0567 0.1921 year#L.AD_new: 2010-0.046 0.2545 -0.3005 -0.0931 0.1307 -0.2238 0.2005 0.185 0.2728 0.1647 0.1968 0.2566 2011 0.0001 0.0008 -0.0008 -0.0025 0.001 -0.0035 0.0574 0.2585 0.2648 0.0617 0.2717 0.2787 2012 -0.0576 0.2272 -0.2848 -0.0431 0.1389 -0.1820 0.0842 0.1885 0.2064 0.0841 0.1993 0.2163 2013 0.0935 0.2524 -0.1589 0.1567 0.1425 0.0141 0.0699 0.1955 0.2076 0.0872 0.195 0.2136 2014 0.0796 0.2445 -0.1649 0.1488 0.1663 -0.0174 0.0845 0.1893 0.2073 0.108 0.1985 0.2260 L.AD_terr -0.1187 -0.0517 -0.0601 0.0978 0.2306 0.1488 0.3329 0.1456 year#L.AD_terr: 20100.4362** 0.3302 0.1060 0.3305 0.0844 0.2461 0.1809 0.2647 0.3206 0.2522 0.2410 0.3488 2011 0.1603 0.1505 0.0098 -0.0615 -0.1101 0.0486 0.2223 0.1659 0.2773 0.3975 0.2119 0.4504 2012 0.0815 0.1027 -0.0213 0.0349 -0.0815 0.1164 0.2317 0.1536 0.2780 0.3342 0.1493 0.3659 2013 0.183 0.0818 0.1012 0.1000 -0.1028 0.2029 0.2318 0.153 0.2777 0.334 0.1481 0.3653 2014 0.0523 0.1592 -0.1070 0.1579 -0.0031 0.1611 0.2345 0.1515 0.2792 0.349 0.1493 0.3795 L.AD_oth 0.0274* 0.0243** 0.0399** 0.0113 0.0143 0.0107 0.0160 0.0096 year#L.AD_oth: 2010 -0.0072 -0.0149 0.0077 -0.0352* -0.0092 -0.0261 0.0167 0.0149 0.0224 0.0198 0.0128 0.0236 2011 -0.0030 -0.0161 0.0131 -0.0284 -0.0044 -0.0240 0.0167 0.0138 0.0217 0.0194 0.012 0.0230 2012 0.0231 0.0092 0.0139 0.01 0.0177 -0.0077 0.0158 0.0131 0.0205 0.0186 0.0115 0.0219 2013 0.0210 0.0177 0.0033 0.0258 0.0422*** -0.0164 0.0162 0.0128 0.0207 0.019 0.0116 0.0223 2014 0.0207 0.0354 -0.0147 0.0315 0.0649*** -0.0333 0.0167 0.0135 0.0215 0.0201 0.0124 0.0236 Other controls As in Table A5.1 N obs/N firms, 1000 37.5 / 9.05 104.8 / 26.8 142.3 / 35.9 37.2 / 9.0 104.2 / 26.8 141.4 / 35.8 R-sq: within 0.0335 0.0287 0.0299 0.0338 0.0300 0.0309 overall 0.0000 0.0024 0.0020 0.0002 0.049 0.0035 Source: Calculation with firm CIT and annual report data. Notes: AD_new, AD_terr and AD_oth refer to AD of new equipment, AD in specially assisted areas and other types of AD, respectively. Otherwise, Notes to Table A5.1 apply. 37 Table A5.4 AD of New Equipment: Effects on Firm Investment Rate and Probability of Investment Dep. Var.: Inv_rate, see (4)-(5) Dep. Var.: 1 if investment ≥ 100 EUR AD = AD_rate, see (6) AD = AD_gain, see (7) T_new C_new D-i-D T_new C_new D-i-D L.AD_new -0.0686 0.0256 0.0307 0.004 0.0894 0.1219 0.0495 0.071 N workers (vs. 50+)#L.AD_new 0 0.3184** 0.4146*** -0.0962 -0.0945 0.1772 -0.2717* 0.1452 0.155 0.2123 0.1069 0.1184 0.1594 1 to 5 0.0595 -0.0373 0.0968 -0.038 0.1036 -0.1416 0.1249 0.1362 0.1848 0.1075 0.0785 0.1331 6 to 10 0.7773*** 0.1444 0.6329** 0.4237*** 0.1017 0.3220** 0.2318 0.142 0.2717 0.1051 0.0976 0.1434 11 to 49 0.2347** -0.0454 0.2801 0.0293 0.0641 -0.0348 0.1019 0.1498 0.1811 0.0602 0.0895 0.1079 L.AD_terr 0.1454*** 0.1247 0.0537*** 0.1512 0.0412 0.1915 0.0188 0.1702 N workers (vs. 50+)#L.AD_terr 0 -1.6803 -0.3311 -1.3492 -1.250*** 0.1571 -1.407*** 1.6234 0.262 1.6430 0.3144 0.3312 0.4565 1 to 5 -0.1306 0.0058 -0.1364 -0.0821** -0.1049 0.0228 0.0832 0.2045 0.2208 0.0403 0.1732 0.1778 6 to 10 0.051 0.0511 -0.0002 0.1013 -0.1721 0.2735 0.457 0.2005 0.4987 0.2117 0.1749 0.2745 11 to 49 -0.118 0.0221 -0.1401 0.0537 -0.1102 0.1639 0.283 0.2415 0.3719 0.0684 0.1784 0.1911 L.AD_oth 0.0377 0.1568*** 0.0704 0.0747** 0.1054 0.0602 0.045 0.0346 N workers (vs. 50+)#L.AD_oth 0 0.1028 -0.2376*** 0.3404** -0.0473 -0.0564 0.0090 0.1198 0.0711 0.1392 0.0523 0.0397 0.0656 1 to 5 0.0826 -0.1835*** 0.2661** -0.0332 -0.0442 0.0110 0.1065 0.0609 0.1226 0.0456 0.0349 0.0574 6 to 10 0.045 -0.1267** 0.1717 -0.0245 -0.032 0.0074 0.1235 0.0642 0.1391 0.0504 0.0363 0.0621 11 to 49 0.0056 -0.1424** 0.148 -0.0145 -0.0148 0.0003 0.1126 0.0642 0.1296 0.0502 0.0369 0.0623 Other controls As in Table A5.1 N obs/N firms, 1000 24.4 / 5.6 70.3 / 17.7 94.7 / 23.3 24.3 / 5.6 70.5 / 17.8 94.8 / 23.5 R-sq: within 0.3592 0.2416 0.2802 0.0396 0.0308 0.0329 overall 0.0915 0.0510 0.0582 0.0009 0.0062 0.0000 Source: Calculation with firm CIT and annual report data. Notes: AD_new, AD_terr and AD_oth refer to AD of new equipment, AD in specially assisted areas and other types of AD, respectively. Columns “T_new” and “C_new” present estimates of fixed effect models Yis = st + βsZ_ADit-1 + γsXit-1 + ui + εit on groups defined in Table A4.1 (panel “New equipment”). Columns D-i-D present only size-varying coefficients s of interactions of the treatment dummy (T) with lagged AD_rate or AD_gain from model (3) estimated on the pooled (T_new and C_new) sample. Robust standard errors clustered on firms in italics. 38 39