## wpiea2024248

## Source details

**Canonical URL:** [wpiea2024248](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024248.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024248.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024248.pdf.json)

---

### Data sources and sample
- Primary datasets:
  - (i) the annual firm-level survey carried out by Valstybės Duomenų Agentūra (Statistics Lithuania);
  - (ii) product-level data from Lithuanian customs Lietuvos Respublikos Muitinė;
  - (iii) product-level trade data from CEPII’s BACI database.
- Annual Survey of Enterprises:
  - Coverage: annual survey of all Lithuanian firms, 1995-2019; main analysis and summary statistics based on 2000 to 2015.
  - Average macro context: Lithuanian economy exhibited an average GDP growth rate of over 4.5 percent during 1995-2019.
  - Sample construction and restrictions:
    - Exclude enterprises with no continuous entries; exclude firms in financial and insurance activities, agriculture, healthcare, and education.
    - Winsorize firms with revenue below/above the 2nd and 98th percentile.
    - Final sample: 96,299 firms observed over 569,540 firm-year observations from 2000-2015.
  - Data shortcomings:
    - 1995-1999 reporting not compatible with international accounting standards.
    - 2016-2019 many variables missing.
    - 2019 reform shifted social security contributions from companies to employees; effects not represented in the data.
- Lithuania customs data:
  - Firm identifier, product code (HS8 digits, except 2010 HS6 digits), import source, export destination, and volumes.
  - Values in litas before 2012; conversion 3.4528 litas to 1 euro.
  - HS8 aggregated to HS6 and maintained as firm-product(HS6)-source/destination pairs.
  - Analysis focused on 2000-2015 (customs raw coverage 1995-2019 but restricted for consistency).
- BACI data:
  - CEPII’s BACI database in HS6; to harmonize over time all HS6 codes converted to the 1992 version.

### Key sample statistics (2000-2015) — preserve original table values
- Sample basis: 569,540 firm-year observations corresponding to 96,299 firms observed between 2000 and 2015.
- Selected descriptive statistics (Mean, Median, Std. Dev., Small (mean), Large (mean)):
  - Revenue: 1170496.03, 103921.50, 11243539.42, 470017.12, 11115233.09
  - Employment: 19.11, 2, 55132.42, 8.08, 962175.606
  - Export: 1302561.63, 669488568029.59, 474027.425002895.98
  - Exporter products: 10.283, 2, 5.53124
  - Expenditure on Energy: 47177.41, 1660, 1205280.3814943.58504778.47
  - Cost of Sales: 929668.44, 558129563562367302.198913612.00
  - Total Debt: 477288.739881.545285012234234081241
  - log(TFP): 2.55, 2.58, 0.71, 2.60, 2.20
  - Markup: 1.11, 1.07, 0.18, 1.11, 1.12
  - UFCI_EX: 57.79, 16.71, 113.30, 43.65, 116.89
  - △UFCI_EX: 0.03, 0.01, 0.71, 0.02, 0.05
- Notes on variables:
  - Revenue = total sales revenue.
  - Employment = average number of employees within a year.
  - Total debt = sum of current liabilities and long-term debt.
  - Export = total export value at the firm level.
  - Expenditure energy = sum of expenditure on fuel, electricity, and energy.
  - Cost of sales = total costs of goods sold.
  - Log(TFP) and markup estimated from translog production function (Appendix A).

### Firm-level Complexity Index — construction and adjustment (formulas preserved)
- Use product-level and country-level complexity measures from the Atlas of Economic Complexity (Harvard Growth Lab).
- Definitions:
  - FCI_EX_ic,t = ∑_j s^EX_ijc,t Complexity_j,t. (1)
  - FCI_IM_ic,t = ∑_j s^IM_ijc,t Complexity_j,t. (2)
    - s^EX(IM)_ijc,t = export (import) share of product j within firm i’s total export (import) to (from) country c in year t.
  - UFCI_EX_i,t = ∑_c w^EX_c,t FCI_EX_ic,t. (3)
  - UFCI_IM_i,t = ∑_c w^IM_c,t FCI_IM_ic,t. (4)
    - w_c,t = Complexity_c,t / (1/C ∑_c Complexity_c,t) (relative country complexity).
  - Combined:
    - UFCI_EX_i,t = ∑_c w^EX_c,t ∑_j s^EX_ijc,t Complexity_j,t = ∑_c ∑_j s^EX_ijc,t w^EX_c,t Complexity_j,t. (5)
    - UFCI_IM_i,t = ∑_c w^IM_c,t ∑_j s^IM_ijc,t Complexity_j,t = ∑_c ∑_j s^IM_ijc,t w^IM_c,t Complexity_j,t. (6)
- Adjustment:
  - Add the absolute value of min Complexity_j,t to the series so new minimum is zero.

### External demand: construction and temporal pattern
- External demand computed using CEPII’s BACI combined with Lithuanian customs data (details in Section 3).
- Product-level demand growth (Davis–Haltiwanger growth rate):
  - △ED_jt,t−1 = ∑_{d∈Ω_d} s_{djt,t−1} (X_djt − X_djt−1) / (1/2 (X_djt + X_djt−1)), (7)
    - where s_{djt,t−1} ≡ 1/2 ( X_{from LT djt−1} / ∑_l X_{from LT l jt−1} + X_{from LT djt} / ∑_l X_{from LT l jt} ), ∑_d s_{djt,t−1} = 1.
  - Instrument (base-period weights 2003):
    - △Z_jt,t−1 = ∑_{d∈Ω_d} s_{djt}^0 (X_djt − X_djt−1) / (1/2 (X_djt + X_djt−1)), (8)
      - s_{djt}^0 ≡ X_{from LT djt}^0 / ∑_l X_{from LT l jt}^0.
- Temporal evolution (2000-2015):
  - Upward trajectory to the financial crisis; negative growth pronounced in 2009; gradual recovery from 2011; downward trajectory toward 2015.
  - Heterogeneity by size: large firms show more robust growth but larger deceleration during the crisis.
- Firm-level aggregation:
  - △ED_it,t−1 = ∑_{j∈Ω_it,t−1} r_ijt,t−1 △ED_jt,t−1
  - r_ijt,t−1 ≡ 1/2 ( V_ijt−1 / ∑_{h∈Ω_it−1} V_iht−1 + V_ijt / ∑_{h∈Ω_it} V_iht )
  - Base-year instrument: △Z_it,t−1 = ∑_{j∈Ω_it^0} r_ijt^0 △Z_jt,t−1 with r_ijt^0 ≡ V_ijt^0 / ∑_{h∈Ω_it^0} V_iht^0.

### Empirical identification and IV strategy
- Panel IV specification (first-difference / growth form):
  - △UFCI_EX_ikt,t−1 = γ △ED_it,t−1 + x′_ikt β + α_i + χ_t + χ_kt + ε_ikt
  - △ED_it,t−1 instrumented by △Z_it,t−1.
  - Controls: firm fixed effects α_i; time fixed effects χ_t; industry-year fixed effects χ_kt; observables x_ikt (including lagged shares with core EU15 and EU entry dummy).
- Shift-Share IV (SSIV) and equivalence approach:
  - △Z_it,t−1 is a Bartik/shift-share instrument using base-year product shares r_ijt^0 and product demand shocks △Z_jt,t−1.
  - Equivalence (Borusyak et al., 2022) recasts firm-level IV into a product-level IV via residualization; requires quasi-random assignment of conditional product-level shocks, sufficient number of products, weak shock error correlation.
  - Support for exogeneity:
    - Shocks computed from world import demand excluding LT.
    - Nearly 4000 products for LT exports provide cross-product variation.
  - Falsification/balance checks (Appendix C):
    - Regressed six firm-level covariates on normalized shift-share instruments; coefficients reported not statistically significant (e.g., Labor productivity: -37.1950, SE 30.5795; Number of observations = 35,376).

### Baseline results — External demand causes complexity gains
- SSIV first-stage highlights (Table 2 excerpts):
  - △ED coefficient examples: 0.0914** (0.0444); 0.0968** (0.0431); 0.101*** (0.0375).
  - EU × △ED reported values: 0.0647 (0.0424); 0.0649 (0.0424); 0.0733* (0.0376); 0.0731* (0.0375).
  - Observations across columns: 43,574; 43,359; 43,359; 43,359; 40,275; 40,275; 40,275.
  - LM statistic examples: 119.3828; 0.6487; 5.5337; 5.5526; 5.7796; 1.7546; 1.7514.
  - Wald F statistics: 155.693; 124.584; 119.412; 119.493; 103.916; 99.633; 99.629.
- Panel IV first-stage highlights (Table 3 excerpts):
  - △ED coefficient examples: 0.0156** (0.0073); 0.0179** (0.0076); 0.0232*** (0.0077).
  - EU × △ED: 0.00828 (0.0077); 0.0144* (0.0078).
  - Observations across columns: 52,500; 41,585; 41,585; 41,585; 33,267; 33,267; 33,267.
- Interpretation:
  - Higher foreign import demand (△ED) leads to significant increases in firm-level complexity (△UFCI) across SSIV and Panel IV specifications.
  - Controlling for lagged import/export shares to EU15 generally strengthens the △ED coefficient.
  - EU accession interacts positively with △ED in some specifications (mild increase in complexity growth post-2004 for firms more dependent on EU15).

### Complexity reduces energy intensity — main IV results
- Dependent variable: △EIy (Davis–Haltiwanger growth rate of firm-level expenditure on energy over sales).
- Instrumented complexity: △UFCISSIV = predicted △UFCI from SSIV first stage.
- Main coefficients (Table 4 and Appendix A.4):
  - △UFCISSIV: -0.00519*** (0.00161) (col 1); -0.00609*** (0.00186) (col 2); -0.00478** (0.00205) (col 5).
  - EU×△UFCISSIV: -0.00635*** (0.00185) (col 3); -0.00637*** (0.00185) (col 4); -0.00504** (0.00205) (col 6); -0.00503** (0.00205) (col 7).
- Model fit and observations:
  - Observations: 68,728 (col 1); 42,055 (cols 2–4); 33,462 (cols 5–7).
  - R-squared: 0.153 (col 1); 0.134 (cols 2–4); 0.131 (cols 5–7).
- Interpretation:
  - Upgrading (higher firm-level complexity) causally reduces energy intensity.
  - EU accession amplifies this negative relationship.

### Heterogeneity: firm size, age, and trade orientation
- Size heterogeneity (Tables 5–6, A.5–A.6):
  - Small firms (≤ 50 employees) drive most of the baseline effect:
    - △UFCISSIV: -0.00553*** (0.00224) (col 1); -0.00849*** (0.00276) (col 2); -0.00726** (0.00321) (col 5).
    - EU×△UFCISSIV: -0.00895*** (0.00279) (cols 3–4); -0.00795** (0.00324) (cols 6–7).
    - Observations: 53,715 (col 1); 29,691 (cols 2–4); 22,624 (cols 5–7).
  - Large firms (> 50 employees) effects smaller/non-significant:
    - △UFCISSIV: -0.00307 (0.00237) (col 1); non-significant in other columns.
    - Observations: 14,345 (col 1); 11,928 (cols 2–4); 10,472 (cols 5–7).
- Age heterogeneity (Tables A.7–A.10):
  - Young firms (≤ 5 years) show larger negative coefficients (e.g., -0.0135*** (0.0052) in Table A.7 col 1).
  - Old firms (> 20 years) show smaller or non-significant effects (e.g., -0.0098** (0.0049) Table A.10 col 1).
  - Conclusion: size matters more than age; young-and-small firms particularly benefit in energy efficiency from upgrading.
- Trade orientation:
  - Small firms with export mix dominated by EU markets associated with higher energy intensity; association strengthens after EU accession.
  - Large firms: importing more from EU15 associated with higher energy intensity pre-accession; this association weakens or vanishes post-accession.

### Financial constraints and interactions with upgrading
- Fin Ratio = (amortization + interest payments of debt) / firm sales.
- Table 7 highlights:
  - △UFCISSIV: -0.0049*** (0.0017) (col 1); -0.0059*** (0.0019) (col 2); -0.0045** (0.0021) (col 5).
  - Fin Ratio main effects: 0.0152 (0.0112) (col 1); 0.0483 (0.0236) (col 2); 0.0648** (0.0298) (col 5).
  - △UFCISSIV × Fin Ratio: generally statistically insignificant in pooled Table 7 (e.g., 0.0015 (0.0062) col 1; -0.0022 (0.0051) col 5).
- Heterogeneous results (Tables A.13–A.14):
  - Small firms:
    - △UFCI_SSIV × Fin Ratio positive and sometimes significant (e.g., 0.0200** (0.0097) Table A.13 col 1), indicating financial constraints can reduce energy-efficiency gains during upgrading for small firms.
  - Large firms:
    - Fin Ratio positive and significant (worsens energy intensity) but △UFCI_SSIV × Fin Ratio negative and significant (e.g., -0.0137** (0.0063) Table A.14 col 1), suggesting upgrading can mitigate negative effects of financial constraints for large firms.
- Interpretation:
  - Financial constraints matter: access to finance is important to realize energy-efficiency gains from upgrading. Small firms are particularly vulnerable.

### Markup dynamics following upgrading
- Local-projection dynamic analysis (Jordà, 2005) of △Markupikt,t+h responses to predicted △UFCI:
  - All firms: slight initial reduction in markups followed by eventual increase; dynamics not statistically significant for the overall sample.
  - Small firms: temporary reduction in markups after upgrading, then return to baseline (transitory competition losses).
  - Large firms: markups increase after a delay following upgrading (market-power gains).
- Interpretation: heterogeneous dynamic responses — large incumbents can capture market power gains over time; small firms face transient competitive pressures.

### Policy implications and recommendations
- Three priority areas:
  - Support small firms in upgrading:
    - Targeted, sustained support because small firms face transitory tougher competition and require help to survive and capture energy-efficiency gains.
  - Mitigate financial frictions:
    - Improve access to finance and ease debt-servicing burdens to enable investments in energy-efficient technologies, especially for small firms.
  - Promote well-functioning input markets and free trade:
    - Facilitate access to cleaner inputs, technologies, and advanced trade partners to enhance energy-efficiency benefits of upgrading.
- Additional guidance:
  - For small firms: easier access to finance during upgrading to increase survival and energy-efficiency gains.
  - For large firms: facilitate entry into and trade with more advanced trade partners to catalyze upgrading and efficiency improvements.
  - Design policies that are time-bound, transparent, consistent with domestic macroeconomic stability, avoid negative cross-border spillovers, are WTO policy-consistent, and preserve competitive neutrality.
  - Coordinate internationally to align green financing mechanisms with industrial policies across borders to stabilize capital flows and ensure reliable access to inputs and financing for small firms.

_Italic: Source: wpiea2024248 - IMF Working Paper — extracted content units 2.1 Data Sources; 3.2 Change in Firm-level Export Demand; 4.2 Firm-level Complexity and Energy Intensity; 6.1 Small Firms; Appendices and Tables as presented in the source PDF._

### 2.1 Data Sources

### 2.1 Data Sources

### Primary datasets
- Three primary datasets used:
  - (i) the annual firm-level survey carried out by Valstybės Duomenų Agentūra (Statistics Lithuania);
  - (ii) product-level data from Lithuanian customs Lietuvos Respublikos Muitinė;
  - (iii) product-level trade data from CEPII’s BACI database.

### Annual Survey of Enterprises
- Coverage and period:
  - Annual survey of all Lithuanian firms carried out by Statistics Lithuania from 1995-2019.
  - Main analysis and summary statistics based on the period from 2000 to 2015.
- Macroeconomic context:
  - During 1995-2019, the Lithuanian economy exhibited an average GDP growth rate of over 4.5 percent.
- Population and variables:
  - Survey is mandatory for all types of firms in Lithuania, except sole proprietors or associations and public administrative entities.
  - Dataset contains detailed firm-level information including firms’ birth and liquidation dates, employment, sectorial activity, ownership structure, assets, liabilities, equities, value-added, revenues and profits, etc.
- Shortcomings of the raw data:
  - (i) from 1995-1999, the reporting criteria were not compatible with international accounting standards;
  - (ii) from 2016-2019, a significant portion of the above variables are missing.
  - A reform implemented in 2019 altered the structure of labor costs by shifting the responsibility of social security contributions from companies to employees; the effects are not represented in the data.
- Sample construction and restrictions:
  - Exclude enterprises with no continuous entries.
  - Exclude firms in financial and insurance activities, as well as those from agriculture, healthcare, and education.
  - Winsorize firms that have revenue below or above the 2nd and 98th percentile in the sample period.
  - Final sample: 96,299 firms observed over 569,540 firm-year observations from 2000-2015.
- Reference for further description: Constantinescu and Proškutė (2019) (as cited in source).

### Lithuania Customs Data
- Content and harmonization:
  - Customs data include firm identifier, product code (HS8 digits, except 2010, which is HS6 digits), import source, export destination, and their respective volume.
  - Values are in litas before 2012; conversion to euros uses 3.4528 litas to 1 euro.
  - To keep product code consistent, HS8 digits are aggregated to HS6 digits and maintained as firm-product(HS6)-source/destination pairs.
- Time coverage focused on analysis: 2000-2015 (customs raw coverage 1995-2019, but restricted for consistency).

### BACI data
- Use: compute external demand for Lithuanian firms.
- Coverage and harmonization:
  - CEPII’s BACI database contains values of bilateral trade flows in the HS 6-digit product classification.
  - To make product codes comparable across time, all HS6 codes are converted to the 1992 version.

### Key sample statistics and firm attributes (summary over 2000-2015)
- Sample basis: 569,540 firm-year observations corresponding to 96,299 firms observed between 2000 and 2015.
- Selected descriptive statistics (Mean, Median, Std. Dev., Small (mean), Large (mean)):
  - Revenue: 1170496.03, 103921.50, 11243539.42, 470017.12, 11115233.09
  - Employment: 19.11, 2, 55132.42, 8.08, 962175.606
  - Export: 1302561.63, 669488568029.59, 474027.425002895.98 (note: table formatting in source; preserve values exactly as presented)
  - Exporter products: 10.283, 2, 5.53124
  - Expenditure on Energy: 47177.41, 1660, 1205280.3814943.58504778.47
  - Cost of Sales: 929668.44, 558129563562367302.198913612.00 (note: preserve table text verbatim)
  - Total Debt: 477288.739881.545285012234234081241
  - log(TFP): 2.55, 2.58, 0.71, 2.60, 2.20
  - Markup: 1.11, 1.07, 0.18, 1.11, 1.12
  - UFCI_EX: 57.79, 16.71, 113.30, 43.65, 116.89
  - △UFCI_EX: 0.03, 0.01, 0.71, 0.02, 0.05
- Notes on variables:
  - Revenue corresponds to total sales revenue.
  - Employment is the average number of employees within a year.
  - Total debt is the sum of current liabilities and long-term debt.
  - Export refers to the total export value at the firm level.
  - Expenditure energy is the sum of expenditure on fuel, electricity, and energy.
  - Cost of sales refers to the total costs of goods sold.
  - Log(TFP) and markup are estimated based on the translog production function specification following the description in Appendix A.

### Firm-level Complexity Index (construction and adjustments)
- Rationale:
  - Use both product-level and country-level complexity measures from the Atlas of Economic Complexity (Harvard Growth Lab) to capture product sophistication and partner-country complexity.
- Definitions (preserve formulas exactly as in source):
  - FCI_EX_ic,t = ∑_j s^EX_ijc,t Complexity_j,t. (1)
  - FCI_IM_ic,t = ∑_j s^IM_ijc,t Complexity_j,t. (2)
    - where s^EX(IM)_ijc,t stands for the export (import) share of product j within firm i’s total export (import) to (from) country c in a given year t.
  - UFCI_EX_i,t = ∑_c w^EX_c,t FCI_EX_ic,t. (3)
  - UFCI_IM_i,t = ∑_c w^IM_c,t FCI_IM_ic,t. (4)
    - where w_c,t = Complexity_c,t / (1/C ∑_c Complexity_c,t) denotes the relative complexity of country c compared to the average country-level complexity at time t.
  - Combined expressions:
    - UFCI_EX_i,t = ∑_c w^EX_c,t ∑_j s^EX_ijc,t Complexity_j,t = ∑_c ∑_j s^EX_ijc,t w^EX_c,t Complexity_j,t. (5)
    - UFCI_IM_i,t = ∑_c w^IM_c,t ∑_j s^IM_ijc,t Complexity_j,t = ∑_c ∑_j s^IM_ijc,t w^IM_c,t Complexity_j,t. (6)
- Adjustment for index scale:
  - Take the absolute value of the minimum of the complexity index, min Complexity_j,t, and add it up to the whole series of the complexity index. The new minimum measure is zero.

### External demand: construction and temporal pattern
- Construction approach:
  - External demand for Lithuanian firms is computed using CEPII’s BACI data combined with Lithuanian customs data; details provided in Section 3 (empirical strategy).
- Temporal evolution (2000-2015):
  - Observed an upward trajectory leading to the onset of the financial crisis.
  - External demand underwent a period of negative growth, particularly pronounced in 2009.
  - Gradual recovery commencing in 2011 with a resurgence in activity, followed by a subsequent downward trajectory towards the end of the sample period in 2015.
- Heterogeneity by firm size:
  - Large firms generally exhibit more robust growth compared to small firms, but large firms also manifest a more pronounced deceleration during the financial crisis.
- Visualization notes (from figures):
  - Figure 1 plots the average Davis-Haltiwanger (D-H) growth rate of firm-level external demand (panel A) and by firm size (panel B).
  - Small firms: ≤ 50 employees; Large firms: > 50 employees.

### Distributional patterns of firm characteristics (energy intensity, complexity, external demand, debt)
- Main empirical observations (from Figure 2 and accompanying text):
  - Energy intensity per output:
    - Large firms cluster around zero with narrower tails; small firms show a broader spread away from zero.
  - Firm complexity:
    - Distribution pattern similar to energy intensity; differences between small and large firms are significant.
  - External demand:
    - Large firms exhibit a pronounced right tail, indicating higher demand variability.
  - Financial debt:
    - Debt growth concentrated around zero for most firms; small firms show elongated tails indicating greater variability.
  - Kolmogorov–Smirnov statistics reported in each panel indicate statistically significant divergence between small and large firms across all dimensions considered.

### Empirical strategy preview (identification of external demand shocks)
- Strategy overview:
  - Construct weighted average foreign import demand in countries that Lithuania sells to, product by product, leaving LT’s own export to the destinations out of the measure of import demand.
  - Instrument these measures with base-year weighted average shocks using a shift-share design following Borusyak et al. (2022).
  - Robustness checks include a traditional panel IV strategy as in Barrows and Ollivier (2021).
- Change in product-level export demand (definitions and formulas preserved):
  - For an LT exporter producing product j at time t, X_djt is aggregate import of product j into destination d from all countries except LT at time t (based on BACI).
  - Davis-Haltiwanger growth rate of product-level export demand:
    - △ED_jt,t−1 = ∑_{d∈Ω_d} s_{djt,t−1} (X_djt − X_djt−1) / (1/2 (X_djt + X_djt−1)), (7)
      - where s_{djt,t−1} ≡ 1/2 ( X_{from LT djt−1} / ∑_l X_{from LT l jt−1} + X_{from LT djt} / ∑_l X_{from LT l jt} ) and ∑_d s_{djt,t−1} = 1.
    - Note: This growth rate operates similarly to a first difference but preserves observations when the shock switches from 0 to a positive number or vice versa, and takes an extremum value of -2 and 2.
  - Instrument to address endogeneity using base-period LT export weights:
    - △Z_jt,t−1 = ∑_{d∈Ω_d} s_{djt}^0 (X_djt − X_djt−1) / (1/2 (X_djt + X_djt−1)), (8)
      - where s_{djt}^0 ≡ X_{from LT djt}^0 / ∑_l X_{from LT l jt}^0.
    - Base period used for export weights: 2003 (Lithuania entered the EU in May 2004).

*Source: wpiea2024248 - 2.1 Data Sources (PDF chapter content).*

### 3.2 Change in Firm-level Export Demand

### 3.2 Change in Firm-level Export Demand

### Construction of firm-level external demand shocks
- Firm-level external demand growth △ED_it,t−1 is constructed by aggregating product-level demand shocks:
  - △ED_it,t−1 = ∑_{j∈Ω_it,t−1} r_ijt,t−1 △ED_jt,t−1
  - r_ijt,t−1 ≡ 1/2 ( V_ijt−1 / ∑_{h∈Ω_it−1} V_iht−1 + V_ijt / ∑_{h∈Ω_it} V_iht )
  - Ω_it,t−1 is the set of products offered by firm i in years t and t−1.
- V_ijt is calculated using the trade data of LT only.
- Davis–Haltiwanger (D-H) growth rate is used as the symmetric growth measure for outcomes (bounded between -2 and 2) and accommodates entry and exit.

### Base-year weighted foreign demand instruments
- Base-year weighted foreign demand instrument △Z_it,t−1 is constructed as:
  - △Z_it,t−1 = ∑_{j∈Ω_it^0} r_ijt^0 △Z_jt,t−1
  - r_ijt^0 ≡ V_ijt^0 / ∑_{h∈Ω_it^0} V_iht^0 (export sales share of product j in base year t^0)
  - △Z_jt,t−1 is the product-level external demand growth (computed excluding LT).
- Time variation in instruments stems only from variation in world export flows X_djt; product destination weights s_djt^0 and firm-product weights r_ijt^0 are fixed at base-period values.

### Identification based on Panel IV
- Empirical specification (first-difference / growth form):
  - △UFCI_EX_ikt,t−1 = γ △ED_it,t−1 + x′_ikt β + α_i + χ_t + χ_kt + ε_ikt
  - △UFCI_EX_ikt,t−1 denotes the D-H growth rate for the export complexity index of firm i in industry k.
  - △ED_it,t−1 is instrumented by △Z_it,t−1.
- Controls and fixed effects:
  - Firm fixed effect α_i captures firm-specific time-invariant components.
  - Time fixed effects χ_t and industry-year fixed effects χ_kt capture common dynamics and industry-specific trends.
  - Vector x_ikt includes additional observables; baseline includes lagged share of firm-specific imports and exports with core EU15 and an EU entry time dummy.
- Rationale:
  - First-difference specification reduces bias from correlation between non-time-varying firm characteristics and levels of demand shocks.
  - Changes in demand shocks △ED_it,t−1 are less likely correlated with firm observables and unobservables than levels.

### Identification based on the Shift-Share IV (SSIV) and equivalence approach
- The instrument △Z_it,t−1 is a standard Bartik / shift-share instrument: pre-EU product shares r_ijt^0 combined with product-level demand shocks △Z_jt,t−1.
- Concerns addressed:
  - Endogeneity from firms choosing product bundles and destinations: mitigated by fixing base-year shares.
  - Lack of IID across firms: addressed via equivalence result (Borusyak et al., 2022) recasting firm-level IV into a product-level IV using residualization (Frisch–Waugh–Lovell).
  - Requirement for validity in design-based approach: conditional product-level demand shocks must be quasi-randomly assigned (expected value conditional on product-level complexity growth error, exposure shares and observables is constant), enough products, and weak correlation among shock errors.
- Contextual support:
  - Shocks are computed from world import demand for each product excluding LT (independent of LT firm choices).
  - For each product j, multiple LT firms export it; product-level shocks can be treated as randomly assigned across LT exporters.
  - LT exports include close to 4000 products, providing extensive cross-product variation and weak dependence among shocks.
- Additional controls and tests:
  - Rich set of controls for time-invariant firm characteristics, aggregate time, and sector-time variations.
  - Observable controls include lagged trade shares with core EU (imports and exports) and a potential break after EU accession.
  - Falsification tests following Borusyak et al. (2022) are reported in Appendix C (test results not reproduced here).

### Brief discussion of residual endogeneity concerns
- Potential concern: unobserved product-level shocks (e.g., LT consumer preference shifts) correlated with base-year shares could bias estimates.
- Mitigations:
  - Controls for firm fixed effects, year, and sector-year effects capture many possible correlated dynamics.
  - LT is a small open economy with an openness ratio around 150% (imports and exports over GDP), implying the domestic market is a small fraction of total sales for largest exporters.
  - The international demand shocks capture global conditions outside LT firms’ control, supporting a quasi-experimental interpretation.

### Baseline results — External demand and firm-level complexity
- Primary finding: higher external demand growth (△ED) causes a significant increase in firm-level complexity (△UFCI) across SSIV and Panel IV specifications.
- Table 2 (SSIV first-stage highlights):
  - Column (1): △ED coefficient = 0.0914** (standard error 0.0444)
  - Column (2): △ED coefficient = 0.0968** (0.0431)
  - Column (3): △ED coefficient = 0.101*** (0.0375)
  - EU × △ED coefficients reported (columns including interaction): 0.0647 (0.0424), 0.0649 (0.0424), 0.0733* (0.0376), 0.0731* (0.0375)
  - Observations: 43,574; 43,359; 43,359; 43,359; 40,275; 40,275; 40,275 (across columns 1–7)
  - Product IV specifications indicated as Z_product, Z_product, Z_prod×EU, Z_prod×EU, Z_product, Z_prod×EU, Z_prod×EU across columns.
  - LM statistic values: 119.3828; 0.6487; 5.5337; 5.5526; 5.7796; 1.7546; 1.7514
  - Wald F statistic values: 155.693; 124.584; 119.412; 119.493; 103.916; 99.633; 99.629
  - Hansen J statistic: 0 in reported columns
  - Notes: △ED is external demand defined in Equation (4.3); △UFCI is D-H growth rate defined in Equation (4.5); EU is a dummy equal to one after 2004.
- Interpretation of SSIV results:
  - Positive and statistically significant △ED estimates indicate firms facing higher foreign import demand tend to increase complexity.
  - When controlling for lagged import and export shares to EU15 (levels or changes), the positive coefficient on △ED generally increases.
  - EU accession interaction results: controlling for changes (△IMP share and △EXP share to EU15) yields mild positive and significant interaction effects, suggesting EU accession brings a mild increase in growth of firm-level complexity for firms increasing dependence on EU15.
- Table 3 (Panel IV first-stage highlights):
  - Column (1): △ED coefficient = 0.0156** (0.0073)
  - Column (2): △ED coefficient = 0.0179** (0.0076)
  - Column (3): △ED coefficient = 0.0232*** (0.0077)
  - EU × △ED coefficients reported: 0.00828 (0.0077); 0.00813 (0.0077); 0.0144* (0.0078); 0.0145* (0.0078)
  - IMP share_EU15_t−1 coefficients: -0.0098 (0.0108); -0.0087 (0.0108) in specified columns
  - EXP share_EU15_t−1 coefficients: 0.0294* (0.0151); 0.0323** (0.0152)
  - △IMP share_EU15_t−1 coefficients: -0.0019 (0.0089); -0.0017 (0.0088)
  - △EXP share_EU15_t−1 coefficients: 0.0095 (0.0121); 0.0093 (0.0121)
  - EU interaction terms with IMP/EXP shares reported where indicated.
  - Observations across columns: 52,500; 41,585; 41,585; 41,585; 33,267; 33,267; 33,267
  - Method: 2SLS with IV indicated as Z_firm or Both (firm and product instruments) across columns.
  - Notes: △ED and △UFCI definitions as in Table 2; EU dummy equals one after 2004.
- Interpretation of Panel IV results:
  - Panel IV estimates are consistent with SSIV: △ED positively associated with growth in firm complexity.
  - Higher export share to EU15 is associated with higher growth of firm-level complexity in some specifications (columns 2 and 3), but this association disappears after removing potential trends in firms’ import and export shares (columns 5–6).
- Economic mechanisms suggested:
  - Increased foreign demand leads firms to diversify product offerings, engage in global value chains, invest in advanced technologies, and adopt more sophisticated strategies — all contributing to increased organizational and operational complexity.

_Italic: IMF Working Paper — excerpted content unit "3.2 Change in Firm-level Export Demand"._

### 4.2  Firm-level Complexity and Energy Intensity

### 4.2 Firm-level Complexity and Energy Intensity

### Relationship between complexity and energy intensity
- Dependent variable: △EIy — the Davis-Haltiwanger growth rate of firm-level expenditure on energy over sales. Energy expenditure includes the firm’s expenditure on gas, fuel and electricity.
- Instrumented complexity: △UFCISSIV is the predicted value of △UFCI based on SSIV in the first stage.
- Main finding: △UFCISSIV is consistently negatively associated with △EIy, indicating that higher firm-level complexity (technological advancement) is associated with reductions in energy intensity.
- Selected coefficient estimates from Table 4 (Second Stage: Complexity and Energy Intensity), dependent variable △EIy:
  - △UFCISSIV: -0.00519*** (0.00161) in column (1)
  - △UFCISSIV: -0.00609*** (0.00186) in column (2)
  - △UFCISSIV: -0.00478** (0.00205) in column (5)
- EU interaction: EU×△UFCISSIV coefficients are negative and statistically significant, indicating a more pronounced relationship after EU accession (EU dummy takes value one after 2004).
  - EU×△UFCISSIV: -0.00635*** (0.00185) in column (3)
  - EU×△UFCISSIV: -0.00637*** (0.00185) in column (4)
  - EU×△UFCISSIV: -0.00504** (0.00205) in column (6)
  - EU×△UFCISSIV: -0.00503** (0.00205) in column (7)
- Export/import shares: level of lagged import or export share to EU15 does not affect energy intensity in columns 2–4, but changes matter:
  - △EXP shareEU15t−1: 0.0515** (0.0220) in column (5)
  - △EXP shareEU15t−1: 0.0514** (0.0220) in column (6)
  - EU×△EXP shareEU15t−1: 0.0617*** (0.0233) in column (7)
- Observations and fit (Table 4):
  - Observations: 68,728 (col 1); 42,055 (cols 2–4); 33,462 (cols 5–7)
  - R-squared: 0.153 (col 1); 0.134 (cols 2–4); 0.131 (cols 5–7)
- Interpretation: EU membership appears to act as a catalyst (regulatory and market channels) for firms to adopt cleaner production processes, amplifying the negative relationship between complexity and energy intensity.

### Heterogeneity across firms (size and export/import orientation)
- Overall pattern: Most baseline results in Table 4 are driven by small firms; effects for large firms are smaller and less consistently significant.
- Small firms (Table 5: Second Stage for Small Firms; Small = firms with less than 50 employees):
  - △UFCISSIV: -0.00553*** (0.00224) in column (1)
  - △UFCISSIV: -0.00849*** (0.00276) in column (2)
  - △UFCISSIV: -0.00726** (0.00321) in column (5)
  - EU×△UFCISSIV: -0.00895*** (0.00279) in columns (3–4)
  - EU×△UFCISSIV: -0.00795** (0.00324) in columns (6–7)
  - △EXP shareEU15t−1: 0.0668* (0.0356) in columns (5–6)
  - EU×△EXP shareEU15t−1: 0.0740** (0.0374) in column (7)
  - Observations: 53,715 (col 1); 29,691 (cols 2–4); 22,624 (cols 5–7)
  - R-squared: 0.175 (col 1); 0.154 (cols 2–4); 0.149 (cols 5–7)
- Large firms (Table 6: Second Stage for Large Firms; Large = firms with more than 50 employees):
  - △UFCISSIV: -0.00307 (0.00237) in column (1)
  - △UFCISSIV: -0.00222 (0.00256) in column (2)
  - △UFCISSIV: -0.00122 (0.00265) in column (5)
  - EU×△UFCISSIV: -0.00240 (0.00255) in columns (3–4)
  - EU×△UFCISSIV: -0.00132 (0.00264) in columns (6–7)
  - △IMP ShareEU15t−1: 0.0430* (0.0243) in columns (5–6)
  - EU×△EXP shareEU15t−1: 0.0513* (0.0278) in column (7)
  - Observations: 14,345 (col 1); 11,928 (cols 2–4); 10,472 (cols 5–7)
  - R-squared: 0.181 (col 1); 0.179–0.180 (cols 2–4); 0.184 (cols 5–7)
- Import/export perspective:
  - Small firms: export product mix dominated by EU markets associates with higher energy intensity, and this association strengthens after EU accession.
  - Large firms: importing more from EU15 is associated with higher energy intensity pre-accession, but this positive association seems to vanish after EU accession, indicating a potential shift toward more sustainable practices for large firms post-accession.
- Age dimension:
  - Young firms (age under 5) experience stronger improvements in energy efficiency compared to older firms; within young firms the effect is driven mainly by small firms.
  - Overall conclusion: size matters more than age in this analysis.

### Financial constraints, complexity, and energy intensity
- Measure: Fin Ratio = sum of amortization and interest payments of debt over firm sales (captures debt-servicing burden / financial constraint).
- Table 7 (Financial Constraints, Complexity and Energy Intensity), dependent variable △EIy:
  - △UFCISSIV: -0.0049*** (0.0017) in column (1)
  - △UFCISSIV: -0.0059*** (0.0019) in column (2)
  - △UFCISSIV: -0.0045** (0.0021) in column (5)
  - Fin Ratio: 0.0152 (0.0112) in column (1)
  - Fin Ratio: 0.0483 (0.0236) in column (2)
  - Fin Ratio: 0.0648** (0.0298) in column (5)
  - △UFCISSIV × Fin Ratio: coefficients generally statistically insignificant in Table 7 (e.g., 0.0015 (0.0062) in col 1; -0.0022 (0.0051) in col 5)
  - EU×△UFCISSIV: -0.0062*** (0.0019) in columns showing the EU interaction
  - △EXP shareEU15t−1: 0.0517** (0.0220) in relevant columns
  - Observations: 68,070 (col 1); 41,955 (cols 2–4); 33,400 (cols 5–7)
  - R-squared: 0.153 (col 1); 0.136 (cols 2–4); 0.134 (cols 5–7)
- Main insights:
  - Financial constraints are, on average, positively associated with firm-level energy intensity, particularly when controlling for changes in lagged export/import shares to EU15.
  - Interaction in Table 7 is statistically insignificant: increasing complexity does not mitigate the impact of financial constraints on energy intensity in the pooled specification.
  - Heterogeneous effects (Appendix Tables A.13–A.14):
    - Small firms: financial constraints, when interacting with upgrading, significantly reduce energy efficiency during upgrading.
    - Large firms: financial constraints directly worsen energy efficiency, but increasing complexity can mitigate this negative effect; there is no independent effect of upgrading alone for large firms.
- Interpretation: Access to finance is important to realize energy-efficiency gains from upgrading; debt burdens can impede investments in energy-efficient technologies and processes.

### Firm-level complexity and markups (dynamic effects)
- Objective: estimate h-period ahead response of firm-level markup growth to predicted changes in △UFCI using local projections (Jordà, 2005). △Markupikt,t+h is the Davis-Haltiwanger growth rate of firm-level markup.
- Dynamics (summary of Figure 3 results):
  - All firms: initial slight reduction in markups (indicative of tougher competition) followed by an eventual increase; dynamics are not statistically significant for the overall sample.
  - Small firms: temporary reduction in markups after upgrading (tougher competition), then markups return to baseline over time. This transitory reduction suggests smaller firms face heightened competition when producing more complex goods.
  - Large firms: markups increase after a delay, indicating large firms may gain market power over time following upgrading.
- Interpretation: heterogeneity is crucial — average effects mask divergent paths for small vs large firms. Incumbent large firms may capture market power gains from complexity, while small firms face transient competition losses.

### Policy implications and recommendations
- Three major policy priorities distilled from the analysis:
  - Support small firms in their upgrading efforts:
    - Targeted policies and sustained support mechanisms are needed because small firms face transitory tougher competition and require help to survive and reap energy-efficiency gains from upgrading.
  - Mitigate financial frictions that hinder energy-efficiency gains:
    - Addressing financial constraints (improving access to finance, easing debt-servicing burdens) is critical to enable investments in energy-efficient technologies, especially for small firms.
  - Promote well-functioning inputs markets and free trade:
    - Facilitate access to cleaner inputs, technologies, and advanced trade partners to enhance the energy-efficiency benefits of upgrading, with attention to the different channels affecting small and large firms.
- Additional guidance:
  - For small firms: provide easier access to finance during upgrading to increase survival chances and energy-efficiency gains.
  - For large firms: facilitate entry into and trade with more advanced trade partners to catalyze upgrading and potential efficiency improvements.
  - Recognize that EU accession and associated regulatory frameworks have amplified the negative relationship between complexity and energy intensity, suggesting regulatory and market integration can improve environmental outcomes.

*Source: IMF Working Paper — Section 4.2 "Firm-level Complexity and Energy Intensity" (figures, tables, and quoted coefficient estimates as presented in the source PDF).*

### 6.1 Small Firms

### 6.1 Small Firms

### Key findings
- Small firms are key contributors to energy savings from upgrading.
- Due to their small scale and intense competition, small firms see limited profitability gains from upgrading and do not benefit from higher profits (markups).
- The market mechanism alone offers limited incentives for small firms to pursue upgrades.
- Small firms’ upgrading leads to significant reductions in energy intensity when exposed to foreign import demand shocks, reinforcing trade-induced efficiency gains (see broader study results).

### Constraints and barriers
- Small scale and intense competition limit market rewards (markups) for upgrading.
- Struggles with access to capital disproportionately affect smaller businesses, making the cost of capital a significant obstacle to energy-efficient upgrades.
- Limited access to complex intermediate inputs and advanced-market networks constrains upgrading potential.
- Fragmentation or trade barriers can cause costly losses of resource access for small firms.

### Policy recommendations
- Export promotion, particularly measures fostering partnerships with advanced economies, to create market opportunities and linkages for upgrading.
- Improve access to complex intermediate inputs to enable production of more complex products and integration into advanced supply chains.
- Ensure well-functioning input markets—including capital, labor, and R&D investments—to support small firms’ upgrading efforts.
- Implement targeted financial mechanisms to alleviate the disproportionate effect of the cost of capital on small firms and enable energy-efficient upgrades.
- Design policies that are time-bound and transparent, consistent with domestic macroeconomic stability, avoid negative cross-border spillovers, are WTO policy-consistent, and preserve as much as possible competitive neutrality.
- Coordinate internationally to align green financing mechanisms with industrial policies across borders, stabilizing capital flows and ensuring reliable access to inputs and financing for small firms.

### Relation to evidence and policy literature
- These differentiated needs are consistent with the recent evidence inLiu(2019) andIMF policy brief, which emphasize that enabling factors behind successful industrial policy include firm size, export orientation and network linkages.
- The targeted approach to finance aligns with IP findings that promote well-structured financing to reduce gaps among firms and allow smaller firms to pursue energy-efficient upgrades effectively (Liu,2019).

*Source: wpiea2024248 - 6.1 Small Firms*

### References

### References (wpiea2024248)

### Literature themes
- Production function and markup estimation methods:
  - Ackerberg, Daniel A, Kevin Caves, and Garth Frazer, “Identification properties of recent production function estimators,” Econometrica, 2015, 83(6), 2411–2451.
  - Olley, G Steven and Ariel Pakes, “The dynamics of productivity in the telecommunications equipment,” Econometrica, 1996, 64(6), 1263–1297.
  - de Loecker, Jan and Frederic Warzynski, “Markups and firm-level export status,” American Economic Review, 2012, 102(6), 2437–2471.
  - de Ridder, Maarten, Basile Grassi, and Giovanni Morzenti, “The Hitchhiker’s guide to markup estimation,” CEPR Discussion Paper No. 17532, 2022.
  - Ackerberg et al. (2015) cited for identification strategy in production function estimation.

- Global value chains, trade, and environmental outcomes:
  - Antràs, Pol, “De-globalisation? Global value chains in the post-COVID-19 age,” NBER Technical Report, 2020.
  - Akerman, Anders, Rikard Forslid, and Ossian Prane, “Imports and the CO2 Emissions of Firms,” CEPR DP16090, 2021.
  - Forslid, Rikard, Toshihiro Okubo, and Karen Helene Ulltveit-Moe, “Why are firms that export cleaner? International trade, abatement and environmental emissions,” Journal of Environmental Economics and Management, 2018, 91, 166–183.
  - Shapiro, Joseph S, “Trade costs, CO2, and the environment,” American Economic Journal: Economic Policy, 2016, 8(4), 220–254.
  - Freitas et al., “Energy use and exporting: an analysis of Chinese firms,” Journal of Evolutionary Economics, 2023, 33(1), 179–207.

- Firm dynamics, productivity, and reallocation:
  - Melitz, Marc J, “The impact of trade on intra-industry reallocations and aggregate industry productivity,” Econometrica, 2003, 71(6), 1695–1725.
  - Davis, Steven J and John Haltiwanger, “Gross job creation and destruction: Microeconomic evidence and macroeconomic implications,” NBER Macroeconomics Annual, 1990, 5, 123–168.
  - Haltiwanger, John, Ron S Jarmin, and Javier Miranda, “Who creates jobs? Small versus large versus young,” Review of Economics and Statistics, 2013, 95(2), 347–361.
  - Garcia-Louzao and Tarasonis, “Productivity-enhancing reallocation during the Great Recession: evidence from Lithuania,” Oxford Economic Papers, 2023, 75(3), 729–749.

- Green innovation, finance, and policy:
  - Bai, Jennie and Hong Ru, “Carbon Emissions Trading and Environmental Protection: International Evidence,” NBER, 2022.
  - Yu et al., “Demand for green finance: Resolving financing constraints on green innovation in China,” Energy Policy, 2021, 153, 112255.
  - Haas et al., “Managerial and financial barriers to the green transition,” Management Science, 2024.
  - Cecere, Corrocher, and Mancusi, “Financial constraints and public funding of eco-innovation: Empirical evidence from European SMEs,” Small Business Economics, 2020, 54, 285–302.

- Economic complexity and product sophistication:
  - Hidalgo, César A, “Economic complexity theory and applications,” Nature Reviews Physics, 2021, 3(2), 92–113.
  - Hidalgo, Cesar A., Ricardo Hausmann, and Partha Sarathi Dasgupta, “The Building Blocks of Economic Complexity,” PNAS, 2009, 106(26), 10570–10575.
  - Stojkoski, Viktor, Philipp Koch, and César A. Hidalgo, “Multidimensional economic complexity and inclusive green growth,” Communications Earth & Environment, 2023, 4(1), 130.

### Appendix A — Production Function Estimation and Markups
- Markup definition (firm-level):
  - μ_it ≡ e^c_it / α^c_it  (Equation (A.1))
  - α^c_it is the variable input cost share of output: variable input costs over sales.
- Measurement and model setup:
  - Output Q_it measured by firms’ total sales revenue Y_it, deflated by industry-specific gross output deflator; y_it denotes log real sales revenue with measurement error ε_it.
  - Inputs vector x_it = (c_it, l_it, k_it) in logs, where c_it is variable input cost, l_it is wage bill, k_it is fixed tangible assets (real values).
- Production function formulation:
  - Assumed: Q_it = Ω_it ̃F(X_it; θ), productivity ω_it enters additively in logs: y_it = ω_it + ̃f(x_it; θ) + ε_it.
- Estimation strategy:
  - Two-step Olley–Pakes approach with identification strategy of Ackerberg et al. (2015).
  - First stage: assume ω_it is a third-order expansion of inputs h(.), run OLS on y_it = g_t(x_it; θ) + ε_it (Equation (A.2)), compute ω̂_it = ĝ_t − ̃f(x_it; θ).
  - Productivity innovations ξ_it = ω_it − m(ω_it−1) with m(.) assumed a third-order expansion; use moment conditions E[ξ_it(θ) [z_it−1  k_it]] = 0 where z_it−1 includes one-period lagged polynomial terms of c_it and l_it; capital k_it treated as predetermined.
- Functional form and markup recovery:
  - Translog production function (Equation (A.3)):
    - ̃f(x_it; θ) = θ_c c_it + θ_l l_it + θ_k k_it + θ_cc c_it^2 + θ_ll l_it^2 + θ_kk k_it^2 + θ_cl c_it l_it + θ_ck c_it k_it + θ_lk l_it k_it.
  - Estimate θ by GMM separately for each 2-digit industry.
  - Empirical firm-level markups (Equation (A.4)):
    - μ̂_it = (θ̂_c + 2 θ̂_cc c_it + θ̂_cl l_it + θ̂_ck k_it) · ̃Y_it / C_it = ê^c_it ̃α^c_it,
    - where ̃Y_it = exp(y_it − ε̂_it) is measurement-corrected sales and ̃α^c_it = C_it / ̃Y_it.

### Appendix B — Complexity Indices (ECI and PCI)
- Data sources:
  - Two datasets from Harvard Growth Lab: Economic Complexity Index (ECI, country level) and Product Complexity Index (PCI, product level).
  - Indices span more than 25 years, ECI involves more than 100 countries, PCI involves more than 1000 products.
- Definitions and interpretation (taken from Harvard Growth Lab descriptions):
  - ECI ranks countries’ complexity based on diversification and complexity of export basket; higher ECI correlates with higher current income and predicts future growth when complexity exceeds income expectations.
  - PCI ranks the diversity and sophistication of productive know-how required to produce a product; PCI is based on how many other countries can produce the product and the economic complexity of those countries.
- Examples (2021 product-level PCI values reported):
  - Most complex product in 2021: "Photographic plates and film, exposed and developed, other than motion-picture film" — PCI = 2.31 (HS 1992 code 3705).
  - Least complex product in 2021: "Tin ores and concentrates" — PCI = -3.37 (HS 1992 code 1221).
- Note on usage in study:
  - Section 2 constructs dependent variable △UFCI using both ECI and PCI; summary statistics of ECI and PCI not provided in the Appendix due to dataset size.

### Appendix C — Falsification Test (Shift-Share IV orthogonality)
- Purpose:
  - Assess plausibility of conditional quasi-random shock assignment following Borusyak et al. (2022) by testing shock orthogonality in SSIV setting.
- Balance check regressions:
  - Regress six firm-level covariates on normalized shift-share instruments (unit variance), controlling for year and year-industry fixed effects; HS4 digit-clustered exposure-robust standard errors reported.
- Reported coefficients and SEs (Table A.1):
  - Labor productivity: Coefficient = -37.1950, SE = 30.5795
  - Labor share: Coefficient = -0.0009, SE = 0.0037
  - Capital intensity: Coefficient = -0.0818, SE = 0.0537
  - Fraction of part-time workers: Coefficient = -0.0137, SE = 0.0084
  - Profit margin: Coefficient = 0.0008, SE = 0.0117
  - Leverage ratio: Coefficient = -0.0006, SE = 0.0059
  - Number of observations = 35,376
- Conclusion stated:
  - No statistically significant relationship found between these predetermined firm production characteristics and the shift-share instruments.
- Caveat noted:
  - Balance check performed only at firm level (not product level) because product-level information comes from global 6-digit trade flow data from CEPII (exogenous to Lithuania) and firm-level data lack product-level variables.

### Appendix D — Additional Tables: Key reported coefficients and specifications
- Table A.2 & A.3 — First stage (fixing country or product complexity):
  - △ED coefficients:
    - Table A.2 (fixing country complexity): △ED = 0.0948** (SE 0.0454) in column (1); 0.0990** (SE 0.0442) in column (2); 0.0997*** (SE 0.0378) in column (3).
    - EU×△ED reported: 0.0680 (SE 0.0434), 0.0681 (SE 0.0435), 0.0720* (SE 0.0378), 0.0719* (SE 0.0377) across columns.
  - Instrument statistics (example):
    - LM statistic = 119.3828 in column (1) of Table A.2; Wald F statistic = 155.693; Hansen J statistic = 0.
  - Observations reported: 43,574; 43,359; 40,275 depending on specification.

- Table A.4 — Second stage: Complexity and energy intensity (full sample):
  - Dependent variable △EI_cogs (Davis–Haltiwanger growth rate of firm-level expenditure on energy over cost of sales).
  - △UFCI_SSIV coefficients:
    - Column (1): -0.00589*** (SE 0.00164)
    - Column (2): -0.00718*** (SE 0.00187)
    - Column (3): -0.00569*** (SE 0.00207)
  - EU×△UFCI_SSIV coefficients:
    - -0.00744*** (SE 0.00187), -0.00745*** (SE 0.00187), -0.00596*** (SE 0.00207), -0.00595*** (SE 0.00207) across columns.
  - Observations: 67,587; 41,895; 33,376 depending on column.
  - R-squared values: 0.152; 0.134; 0.132 across reported columns.

- Table A.5 — Small firms (Firm Size: Small < 50 employees):
  - △UFCI_SSIV coefficients for △EI_cogs:
    - -0.00653*** (SE 0.00228) in column (1)
    - -0.0102*** (SE 0.00279) in column (2)
    - -0.00861*** (SE 0.00326) in column (3)
  - EU×△UFCI_SSIV:
    - -0.0106*** (SE 0.00281), -0.0106*** (SE 0.00282), -0.00934*** (SE 0.00329), -0.00934*** (SE 0.00329) across columns.
  - Observations: 52,660; 29,573; 22,570 depending on column.
  - R-squared: 0.173; 0.153; 0.150.

- Table A.6 — Large firms (Firm Size: Large > 50 employees):
  - △UFCI_SSIV coefficients for △EI_cogs:
    - -0.00345 (SE 0.00239), -0.00242 (SE 0.00257), -0.00160 (SE 0.00265) across columns (non-significant).
  - EU×△UFCI_SSIV similarly non-significant in reported columns.
  - Observations: 14,271; 11,888; 10,439.
  - R-squared values around 0.179–0.183.

- Tables A.7–A.12 — Heterogeneity by age and size:
  - Young firms (≤ 5 years) show larger negative coefficients:
    - Table A.7 (△EI_y): △UFCI_SSIV = -0.0135*** (SE 0.0052) in column (1); -0.0234*** (SE 0.0086) in column (2); -0.0229* (SE 0.0138) in column (3).
    - EU×△UFCI_SSIV = -0.0226*** (SE 0.0088) and similar values across columns.
  - Old firms (> 20 years) exhibit smaller or non-significant effects (Table A.10):
    - △UFCI_SSIV = -0.0098** (SE 0.0049) in column (1); other columns show -0.0085 (SE 0.0054), -0.0071 (SE 0.0055) (some not significant).
  - Combinations (Young & Small, Young & Large, Old & Small, Old & Large) reported with corresponding sample sizes, coefficients, and R-squared values (see tables A.8–A.12 for exact cells).

- Tables A.13 & A.14 — Financial constraints interaction (Fin Ratio = amortization + interest over sales):
  - Small firms (Table A.13):
    - △UFCI_SSIV = -0.0060*** (SE 0.0023) in column (1).
    - Fin Ratio main effect often small/non-significant in columns (e.g., 0.0086 (SE 0.0125) in column (1)).
    - △UFCI_SSIV × Fin Ratio = 0.0200** (SE 0.0097) in column (1); several specifications report positive and sometimes significant interactions (e.g., 0.0478** (SE 0.0208) in later columns).
    - Observations: 53,114; 29,615; 22,584 across columns.
  - Large firms (Table A.14):
    - △UFCI_SSIV not significant (e.g., -0.0016 (SE 0.0025)).
    - Fin Ratio positive and significant in many columns (e.g., 0.0885*** (SE 0.0255) in column (1); 0.0904** (SE 0.0403) in column (2)).
    - △UFCI_SSIV × Fin Ratio negative and significant: -0.0137** (SE 0.0063) in column (1); -0.0105** (SE 0.0042) in column (2); consistent negative interaction across reported columns.
    - Observations: 14,303; 11,904; 10,449 depending on column.
- Notes on table reporting conventions:
  - Standard errors are clustered robust and reported in parentheses.
  - Significance notation: ***p<0.01, **p<0.05, *p<0.1.
  - EU is a dummy variable that takes value one after 2004 in specified regressions.
  - △UFCI_SSIV denotes predicted value of △UFCI based on SSIV first stage.
  - △EI_cogs and △EI_y denote Davis–Haltiwanger growth rates of energy expenditures over cost of sales or over sales, respectively, depending on table.

*Content extracted from wpiea2024248 - References (IMF Working Paper: The Heterogeneous Impacts of Firm Upgrading on Energy Intensity, Working Paper No. WP/2024/248).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024248.pdf_