## wp18272

## Source details

**Canonical URL:** [wp18272](https://www.imf.org/-/media/files/publications/wp/2018/wp18272.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18272.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18272.pdf.json)

---

### Introduction and motivation
- Paper title: Economic Policy Uncertainty in Turkey.
- Authors: La-Bhus Fah Jirasavetakul and Antonio Spilimbergo.
- Authorized for distribution by Donal McGettigan and Krishna Srinivasan.
- Date: December 2018.
- JEL Classification Numbers: D25, D80, E22, E66, L20, M51.
- Keywords: Policy uncertainty, economic uncertainty, Turkey, firm-level, sector-level, investment decisions, employment growth, leverage strategies, diff-in-diff estimation.
- Core purpose:
  - Construct a news-based economic policy uncertainty (EPU) index for Turkey (monthly index from 1996 to 2017).
  - Assess how EPU affects Turkish firms (investment, employment, leverage).
  - Address endogeneity and reverse causality using a difference-in-differences approach exploiting firm/sector exposure via share of irreversible investment.

### EPU index construction and properties
- Data sources and search criteria:
  - Primary news source: Dow Jones’ Factiva Top Sources and English-language Turkish news sources (Factiva coverage nearly 33,000 news sources).
  - Inclusion: articles must mention words in each of three areas—economics, policy, and uncertainty—and one of the three mentions must be within 10 words of “Turkey” or “Turkish”; filtered by geographic coverage tagged as related to Turkey; articles must contain more than 99 words and be tagged as related to economics and politics.
  - Policy terms include Treasury, Ministry of Finance, Central Bank of Turkey, Banking Regulation and Supervision Agency and variants.
- Normalization and scaling:
  - Monthly raw counts normalized by total counts of economics- and politics-related articles in the specified news base containing the term “today”.
  - Alternative normalizers tested did not materially affect the EPU index.
  - Scaled EPU index standardized to a one-unit standard deviation and normalized to a mean of 100 (1996–2017).
  - Annual and quarterly indices created by aggregating monthly counts and similarly standardized.
- Evolution and correlations:
  - Major spikes correspond to: 2001 Turkish economic crisis; 2002 general elections; 2003 crisis in Turkish-American relations; 2011 general elections; 2013 Gezi Park protest; large depreciation in 2014; 2015 sanctions by Russia and general elections; 2016 Istanbul Airport bombing and the failed coup attempt.
  - Correlation of monthly news-based EPU index with CDS spread: 0.36.
  - Correlation increases to around 0.5 when using annual and quarterly EPU indices with annual and quarterly average CDS spreads.
  - Notable discrepancies: EPU spiked in 2013 (Gezi protests) while CDS spread remained subdued; EPU spiked around the failed coup attempt while CDS spread did not increase.

### Data, variables, and empirical framework
- Sectoral Company Accounts data:
  - Source: Company Accounts statistics compiled by the Central Bank of the Republic of Turkey (CBRT), sector-level aggregates for 1999–2015.
  - Sector classification: NACE Rev.1.1 (1999–2009); NACE Rev.2 (2010 onwards).
  - Panel covers a 17-year period (1999–2015) with around 27 sectors comparable over time.
  - Number of firms in the survey: around 7,000 firms in 1999 to around 10,000 firms in 2015.
- Main sectoral variables:
  - Real investment rate: ratio of real (net) investment to real stock of capital at beginning of year (deflated by GFCF deflator).
  - Employment growth: simple annual (net) growth rate of employment.
  - Net borrowing variables: total, short-term, and long-term net borrowings (changes in liabilities), calculated in US dollars and normalized by levels at the beginning of the year.
  - Investment irreversibility: proxied by share of machinery, plant, and equipment, and intangible fixed assets to total assets.
  - Controls: sales growth in real terms, operating cashflows, real borrowing costs, debt to total asset ratios, Real GDP growth and inflation and their one-year lags, one-year ahead GDP forecast, CDS spread, European EPU index.
- Baseline reduced-form specification (symbolic form preserved):
  - Y_{s,t} = α + γ U_{t−l}^Q + β1 Z_{s,t−1} + β2 X_{s,t} + β3 K_t + θ_s + λ_t^* + ε_{s,t}
    - Y_{s,t}: outcome in sector s and year t (investment rate, employment growth, net borrowing).
    - U_{t−l}^Q: quarterly EPU index (Q ∈ {Q1,Q2,Q3,Q4}), with l ∈ {0,1}.
    - θ_s: sector fixed effect; λ_t^*: time trend.
- Identification strategy:
  - Difference-in-differences (DID) exploiting heterogeneity in sector exposure measured by investment irreversibility (treated as continuous treatment intensity T_s).
  - Coefficient of interest ψ̂ measures impact of a one standard deviation increase in quarterly EPU on sector outcomes.
  - T_s and sector fixed effects θ_s are perfectly collinear, so T_s is not included separately.

### Main empirical findings — investment
- Baseline (no DID interaction):
  - Using Q1_t EPU index, coefficient γ nearly -4.882** in one specification indicates a one standard deviation increase in first-quarter EPU associated with a 5-percentage point lower real annual investment rate in the same year (statistically significant at the 90 percent level in reported column).
  - Higher operating cashflows and sales growth associated with higher investment; higher real borrowing costs associated with lower investment.
- DID results (investment irreversibility as treatment intensity):
  - DID negative impact magnitudes "around 9 to 13 percent" (reported coefficients EPU × Inv. irreversibility: -12.767**, -9.444*, -6.862**, -6.001* in selected specifications).
  - Interpretation: a one standard deviation increase in quarterly EPU around end of earlier year or beginning of current year induces firms in high-irreversibility sectors to reduce real annual investment rate by around 9 to 13 percentage points, holding other things constant.
  - Recent years: EPU index increased by about 0.2 to 0.5 per year, implying a decline in the real investment rate for firms in these sectors by about 2 to 6.5 percentage points (context: mean and median real investment around 8 to 10 percent, respectively, for 2015).
- Alternative dependent variable (investment as share of total assets):
  - Baseline impact smaller and not statistically significant in some columns.
  - DID results remain statistically significant; one standard deviation increase in quarterly EPU could reduce investment-to-total-asset ratio by around 6 percentage points in current year (selected DID coefficients: -6.862**, -6.001*; linear combination examples: -5.988** (2.335)).
- Timing evidence:
  - Weaker but statistically significant negative impact of policy uncertainty in Q3 of previous year on current-year real investment rate, suggesting investment decisions could be made as early as second half of previous year.

### Main empirical findings — employment
- Baseline employment results sometimes positive (reported EPU index coefficients: 4.207, 8.358**, 5.595**, 9.570** in selected columns).
- DID interaction (EPU × Inv. irreversibility) turns estimated impact negative but not statistically significant (examples: -3.341, -2.969).
- Interpretation caveats:
  - Employment measure reflects aggregate number of employees only; does not capture composition (skilled vs low-skilled), contract types, or capital-skill complementarities.
  - Turkey-specific labour supply growth: labour force grew at an average of 3.5 percent per annum during the past decade, about 4 percent per annum during 2010-2011, and above 6 percent in 2014 when policy uncertainty picked up—these structural shifts could mute employment impacts.
  - Firms postponing investment may substitute labour for capital in short term, producing ambiguous aggregate employment response.

### Main empirical findings — leverage (net borrowing)
- Baseline regressions show EPU associated with higher net borrowing (examples of positive EPU index coefficients for borrowing outcomes: 14.078***, 18.170***, 18.822***, 27.388*** in appendix tables).
  - Potential interpretations: firms build cash buffers; possible reverse causality if central bank policy uncertainty rises in overheating periods.
- DID estimates show negative and statistically significant interaction coefficients:
  - A one standard deviation increase in EPU around end of previous year and beginning of current year leads to around 17 to 21 percentage points reduction in ratio of total net borrowing to total liabilities in current year (selected EPU × Inv. irreversibility interactions: -21.064*** (6.735), -20.934*** (6.654), -19.665** (7.745)).
- Maturity breakdown:
  - Reduction in total net borrowing driven by reduction in net short-term borrowing (examples: L.EPU x Inv. irreversibility in short-term borrowing panels: -19.778** (7.931); linear combinations indicate subdued short-term borrowing).
  - Results for long-term borrowing smaller and often insignificant when using EPU of last quarter of preceding year.
  - Implication: EPU leads to change in debt-term structure, particularly reducing short-term liabilities.

### Robustness checks and alternative specifications
- Alternative treatment intensity variables:
  - MPE intensity (machinery, plant, equipment share) yields similar negative interactions: L.EPU x MPE intensity reported examples -12.096** (5.329); -21.323*** (6.688).
  - Depreciation rate (inverse treatment intensity) yields positive interaction coefficients (examples: L.EPU x Depreciation rate: 16.575** (6.083); 24.696*** (8.574)), consistent with interpretation that firms less exposed cut less.
  - Specific reported magnitudes: a one standard deviation increase in EPU at beginning of year results in a 12-percentage point decline in investment rate and around 20 percentage points lower net short-term borrowing in some specifications.
- Time-dummy specification:
  - Replacing linear/quadratic time trends with time-dummies yields generally robust results (examples: L.EPU x Inv. irreversibility: -12.676** (5.163); -6.808** (3.030); -20.934*** (6.654) in Table A.4).
- Sector exclusion checks:
  - Excluding construction, energy, education, and health care sectors leaves main results intact (examples repeated: L.EPU x Inv. irreversibility: -12.676** (5.163); -6.808** (3.030); -20.934*** (6.654)).
- Annualized EPU regressions and lags:
  - Using annualized EPU with various quarter lags shows statistically significant negative interactions for Nq_Lagged_EPU = 1, 2, 3 in various specifications (examples: Nq_Lagged_EPU = 2 — L.EPU x Inv. irreversibility: -9.400* (4.620); Nq_Lagged_EPU = 3 — -10.554** (4.410); Nq_Lagged_EPU = 3 — -18.013*** (5.956); Nq_Lagged_EPU = 4 — -18.832** (8.015)).
  - Finding: collective information on policy uncertainty from third quarter last year onwards could influence current-year investment and net borrowing.
  - Caution: contemporaneous annual impacts likely subject to reverse causality; reverse causality weakens but remains for longer lags—motivating quarterly EPU use in main specifications.
- Other robustness notes:
  - Observations in panel regressions: 383 (all main specifications).
  - R2 reported across specifications (examples): 0.503, 0.505, 0.510, 0.508, 0.521, 0.524, 0.526, 0.527 (Table 1 selections); appendix tables report R2 values across robustness checks (examples: 0.510; 0.526; 0.555; 0.527; 0.365; 0.509; 0.538; 0.561; etc.).
  - Significance legend preserved: * p<0.1; ** p<0.05; and *** p<0.01.
  - Standard errors clustered at sector level in all regressions.

### Key conclusions and policy implications
- Empirical summary:
  - An increase in the first-quarter EPU index by one standard deviation is associated with a 5-percentage point lower real annual investment rate in the same year; the magnitude more than doubles for sectors with high irreversibility (DID estimates around 9 to 13 percentage points, with selected interaction coefficients such as -12.767**, -9.444*).
  - Employment effects are ambiguous: baseline estimates sometimes positive, DID estimates negative but statistically insignificant; compositional and labour-supply factors (labour force growth: 3.5 percent per annum average over past decade; about 4 percent per annum during 2010-2011; above 6 percent in 2014) may mute observable employment responses.
  - Sectors with high irreversibility reduce short-term borrowing in periods of uncertainty; DID interactions for borrowing include large-magnitude coefficients (examples: -21.064*** (6.735); -19.778** (7.931)).
- Policy implication:
  - Policy makers should strive to reduce policy uncertainty given measurable negative effects on investment and on the term structure of corporate borrowing.
  - Authors recommend extending similar sectoral analysis to other emerging markets where uncertainty remains high.

*Source: WP/18/272 Economic Policy Uncertainty in Turkey (Sections I–V and Appendix, December 2018).*

### Section 1

### Economic Policy Uncertainty in Turkey

### Introduction
- Paper title: Economic Policy Uncertainty in Turkey
- Authors: La-Bhus Fah Jirasavetakul and Antonio Spilimbergo
- Authorized for distribution by Donal McGettigan and Krishna Srinivasan
- Date: December 2018
- JEL Classification Numbers: D25, D80, E22, E66, L20, M51
- Keywords: Policy uncertainty, economic uncertainty, Turkey, firm-level, sector-level, investment decisions, employment growth, leverage strategies, diff-in-diff estimation
- Abstract core purpose:
  - Construct a news-based economic policy uncertainty (EPU) index for Turkey.
  - Assess how EPU affects Turkish firms.
  - Address endogeneity and reverse causality using a difference-in-differences approach exploiting firm/sector exposure via share of irreversible investment.

### Motivation and framing
- Uncertainty leads economic agents to postpone consumption and investment (citations: Arrow, 1968; Bernanke, 1983; Pindyck, 1988; Rodrik, 1991).
- Policy uncertainty has been argued to contribute to slow recovery from the global economic crisis and has been heightened by political turmoil.
- Turkey is an ideal case study given recent events that increased policy uncertainty (failed coup attempt, dismissals/detentions/suspensions of many public employees, constitutional referendum to replace parliamentary system with presidential system).

### Challenges in studying EPU in Turkey
- Two main challenges identified:
  - Lack of a publicly available and unbiased index of EPU for Turkey covering a long period until recent years.
  - Reverse causality: policy uncertainty affects macro variables and macro shocks can generate policy uncertainty, complicating causal interpretation.

### Approach and identification
- EPU index construction:
  - News-based approach following Baker et al (2016).
  - Monthly index constructed from 1996 to 2017.
  - Main difference from many country EPU indices: primarily uses foreign newspapers rather than Turkish newspapers to capture international perceptions and to access longer spans of coverage.
- Identification strategy:
  - Difference-in-differences (DID) exploiting heterogeneous exposure to EPU across firms/sectors based on degree of irreversible investment.
  - Rationale: firms with greater irreversible investments (higher upfront fixed and sunk costs) derive greater value from waiting and are more intensely "treated" by increases in EPU.
- Outcomes examined at the firm/sector level:
  - Fixed investment decisions.
  - Employment decisions.
  - Leverage strategies.

### Key empirical findings (as reported)
- Causal effects:
  - For sectors with high exposure to economic policy uncertainty:
    - An increase in economic policy uncertainty lowers investment.
    - An increase in economic policy uncertainty lowers debt leverage.
  - Deleveraging effect is driven more by a reduction in short-term borrowing, affecting the term structure of debt issuance.
- Employment effects:
  - Results offer limited guidance about impacts of policy uncertainty on employment.
  - Possible explanations:
    - Effects on the composition of employment (e.g., skilled versus low-skilled, short-term versus long-term contracts) rather than overall levels.
    - Substantial increase in labour supply over the period and structural changes in the country’s growth dynamics and drivers could have muted impacts.
- Robustness:
  - Results robust to alternative measures of exposure to policy uncertainty and more restrictive specifications with a full set of time controls.

### Relation to existing literature and contribution
- Theoretical background:
  - Real-options approach: ability to delay irreversible investment is akin to holding an option; higher uncertainty increases option value of waiting and thus discourages investment (references include Dixit and Pindyck, 1994; Bernanke, 1983; Pindyck, 1988).
  - Investment responses operate on intensive and extensive margins when multiple capital goods are present.
- Distinguishing features of this paper:
  - Uses exposure defined by investment irreversibility to identify causal effects of EPU.
  - Uses foreign (international) news sources to construct an EPU index tailored to international perceptions, argued to be particularly relevant for Turkey given its reliance on external conditions and sentiments.
  - Addresses endogeneity concerns via DID exploiting heterogeneity in irreversibility rather than relying on instruments whose validity can be questioned.

### Measuring Economic Policy Uncertainty — data sources and construction details
- News-based EPU index rationale:
  - Captures frequency of news articles about economic policy uncertainty, reflecting public perception and anticipation of volatility (including non-realized anticipations).
  - Tailorable to policy-specific uncertainty beyond financial-market volatility measures.
- Primary news source and coverage:
  - Primary news source: Dow Jones’ Factiva news aggregator.
  - Factiva coverage: nearly 33,000 news sources, both online and print, in 28 languages from nearly 200 countries since 1980s.
  - Authors confine the news database to English-language Factiva Top Sources and Turkish news sources to capture international perceptions relevant to Turkey.
- Advantages over traditional measures:
  - Traditional measures (macroeconomic variable volatility, forecaster dispersion, market volatility, CDS spreads) may reflect realized uncertainty, insider financial views, or broad macro uncertainty rather than policy-specific or anticipatory uncertainty.
  - News-based measure captures anticipations and policy-specific discussions.

*Source: WP/18/272 Economic Policy Uncertainty in Turkey, Section I–III (December 2018).*

### Section 2

### wp18272 - Section 2

### Construction of Turkey’s Economic Policy Uncertainty (EPU) Index
- News-article selection criteria:
  - Articles must mention words in each of three areas: economics, policy, and uncertainty.
  - The three mentions must be within 10 words of “Turkey” or “Turkish”.
  - Searches are filtered by geographic coverage to include only articles tagged by Factiva as related to Turkey.
  - Policy terms include fiscal and monetary policies and relevant institutions (Treasury, the Ministry of Finance, the Central Bank of Turkey, the Banking Regulation and Supervision Agency, and their variants and abbreviations).
  - An eye-ball audit of articles was conducted to identify wordings improving search precision; Appendix 1 lists the wordings for each keyword.
- Normalization and scaling:
  - Monthly raw counts of EPU-related articles are normalized by counts of all news articles to account for variation in overall article volume.
  - The normalizer is the total counts of economics- and politics-related articles in Factiva Top Sources and the English-language Turkish news sources containing the term “today” (due to search platform limitations).
  - Alternative normalizers tested: (i) total counts of all “today”-articles in Factiva’s Top Sources and English-language Turkish sources; and (ii) total counts of non-sports-related “today”-articles in Factiva’s Top Sources and English-language Turkish sources. Different normalizers do not materially affect the EPU index.
  - For ease of interpretation, the scaled EPU index is standardized to a one-unit standard deviation and normalized to a mean of 100 (from 1996 to 2017).
- Temporal aggregations:
  - Monthly EPU index constructed from monthly article counts and normalizers.
  - Annual EPU index: scale 12-month raw counts of identified articles by 12-month counts of the normalizer; standardized to a unit standard deviation with mean 100.
  - Quarterly EPU index: computed using respective 3-month counts and similar standardization.

### Evolution and properties of the constructed EPU index for Turkey
- Major spikes in the EPU index correspond to:
  - 2001 Turkish economic crisis
  - 2002 general elections (AKP coming to power)
  - 2003 crisis in Turkish-American relations
  - 2011 general elections
  - 2013 Gezi Park protest
  - large depreciation in 2014
  - 2015 sanctions by Russia and general elections
  - 2016 Istanbul Airport bombing and the failed coup attempt
- Other smaller spikes noted during:
  - onset of the EU accession negotiations in 2005
  - 2010 referendum on constitutional reform
- Relation to financial market indicators:
  - Monthly news-based EPU index correlation with CDS spread: 0.36.
  - Correlation increases to around 0.5 when using annual and quarterly EPU indices with annual and quarterly average CDS spreads.
  - Illustrative discrepancies:
    - EPU spiked in 2013 (Gezi protests) while CDS spread remained subdued.
    - EPU spiked around the failed coup attempt while CDS spread did not increase.
- Interpretation and robustness:
  - Lower relative EPU during the Global Financial Crisis of 2007-2009 is explained by a large global increase in article volume (normalizer) driven by US and European crisis coverage; thus the index reflects Turkey’s perceived policy uncertainty relative to other economies in the selected English-language news base.
  - The constructed index is considered appropriate for research questions focusing on relative policy uncertainty facing domestic enterprises in an increasingly globalized economy.

### Company Accounts Data and key sectoral variables (1999–2015)
- Data source and coverage:
  - Sector-level data from 1999 to 2015 using aggregate sectoral balance sheets, income statements, and non-financial enterprise information from the Company Accounts statistics compiled by the Central Bank of the Republic of Turkey (CBRT).
  - Sector classification: NACE Rev.1.1 from 1999 to 2009; NACE Rev.2 from 2010 onwards.
  - Sector-level panel: 17-year period (1999-2015) with around 27 sectors of production comparable over time.
  - Number of firms in the survey: ranges from around 7,000 firms in 1999 to around 10,000 firms in 2015.
  - Data checking: all data are checked for consistency; only data passing checks and/or corrected are included in the final database for aggregation.
- Strengths and limitations:
  - Strength: substantially larger coverage than firm-level vendor databases for Turkey (e.g., Bureau Van Dijk coverage below 100 firms in 1990s to around 100-300 firms in 2010s; Bloomberg covers around 300-400 companies but with limited non-financial information).
  - Limitation: cannot capture firm-level heterogeneity and lacks higher-frequency panel granularity.
- Main sectoral variables used:
  - Real investment rate: ratio of real (net) investment to real stock of capital at beginning of year; net investment and capital stock deflated by the gross fixed capital formation (GFCF) deflator.
  - Employment growth: simple annual (net) growth rate of employment.
  - Net borrowing variables: total, short-term, and long-term net borrowings (changes in liabilities), calculated in US dollars and normalized by levels at the beginning of the year.
  - Investment irreversibility: proxied by share of machinery, plant, and equipment, and intangible fixed assets (including know-how, goodwill, R&D) to total assets.
  - Controls available: sales growth in real terms, operating cashflows, real borrowing costs, debt to total asset ratios.
- Summary statistics and sectoral patterns:
  - Overall investment rate average over 2000-2015: around 8 percent.
  - Investment rate was negative in 2001 and in the early 2010s, coinciding with rising EPU.
  - Employment per firm and net borrowing roughly followed same trend during 2000s; employment per firm peaked in 2010-2011 then deteriorated.
  - Total net borrowing largely moved with net short-term borrowing; correlation declined over time as short-term liabilities’ share decreased.
  - Investment irreversibility by sector: transport, manufacturing, mining, and health sectors averaged >50 percent irreversible assets; trade, construction, and renting and leasing sectors around or less than 10 percent irreversible assets.

### Other data sources used as controls
- Official macroeconomic statistics: GDP and inflations.
- GDP forecasts: Consensus Economics, Inc.
- External policy uncertainty control: European EPU index from Baker et al (2016).
- Financial market data: CDS spread and volatility of Turkey’s stock price index from Bloomberg.

### Empirical framework
- Objective: Investigate whether an increase in EPU impacts business decisions—investment, employment, and net borrowing—using a reduced-form firm-level investment model adapted to sector-year data.
- Baseline regression specification (symbolic form preserved):
  - Y_{s,t} = α + γ U_{t−l}^Q + β1 Z_{s,t−1} + β2 X_{s,t} + β3 K_t + θ_s + λ_t^* + ε_{s,t}
    - Y_{s,t}: outcome in sector s and year t (investment rate, employment growth, net borrowing).
    - Z_{s,t−1}, X_{s,t}: vectors of time-varying sector aggregates (sales growth, operating cashflows, real borrowing costs, debt-to-total asset ratios).
    - K_t: time-varying common controls (real GDP growth and inflation and their one-year lags, one-year ahead GDP forecast, CDS spread, European EPU index).
    - θ_s: sector fixed effect.
    - λ_t^*: time trend.
- Use of quarterly EPU in sector-year regressions:
  - Quarterly EPU index U_t^Q (Q ∈ {Q1,Q2,Q3,Q4}) used to capture contemporaneous and lagged impacts; l ∈ {0,1} denotes annual lag of the quarterly EPU index relative to year t.
  - Two main specifications: U_{t}^{Q1} (first quarter of current year) and U_{t−1}^{Q4} (last quarter of previous year), reflecting assumption that final investment decisions are made at year-beginning or end of previous year; impacts of other quarters are also tested.
- Addressing endogeneity:
  - Time-varying controls included to proxy for omitted variables correlated with outcomes and EPU (economic and financial market conditions and prospects).
  - Additional strategies described (beginning of a Difference-in-Differences specification) to mitigate remaining endogeneity concerns.

*Source: wp18272 - Section 2 (pdf: wp18272 - Section 2).*

### Section 3

### wp18272 - Section 3

### Endogeneity, Reverse Causality, and Identification Strategy
- The conditional correlation between outcome variables and the EPU index may be driven by omitted variables that influence both outcomes and policy uncertainty; reverse causality is also a concern.
- To mitigate endogeneity and reverse causality concerns, the paper adopts a difference-in-differences (DID) technique.
- Treatment sectors: sectors with high exposure to EPU (treated more intensely by policy uncertainty).
- Control sectors: sectors with low exposure to policy uncertainty.
- DID framework assumption: conditional on controls and in the absence of policy uncertainty, treatment and control sectors evolve similarly; under these assumptions DID can provide a causal interpretation of the estimated impact of policy uncertainty on the treatment group relative to the counterfactual deduced from the control group.
- Investment irreversibility measure: share of machinery, plant, and equipment, and intangible fixed assets to total assets; treated as a continuous treatment intensity variable (푇푠) in a panel DID regression.
- Coefficient of interest 휓̂ measures the impact of a one standard deviation increase in the quarterly EPU index on investment rate, employment growth, and net borrowing of firms.
- Note: 푇푠 and sector fixed effects 휃푠 are perfectly collinear so 푇푠 is not included separately.

### Investment Decisions (Results and Magnitudes)
- Estimation details:
  - Equations (1) and (2) estimated using two choices of quarterly EPU index: EPU index of the first quarter of the current year (Q1 t) and EPU index of the last quarter of the preceding year (Q4 t-1).
  - Sector fixed effects included; linear and quadratic time trends included.
  - Standard errors clustered at the sector level.
- Baseline (no DID treatment interaction):
  - Column (1): Using Q1 t EPU index, estimate of 훾 is statistically significant at the 90 percent level and economically significant.
  - A coefficient of nearly 5 (Column (1)) indicates that an increase in the first-quarter EPU index by one standard deviation is associated with a 5-percentage point lower real annual investment rate in the same year.
  - Higher operating cashflows and sales growth associated with higher investment; higher real borrowing costs significantly correlate with lower investment.
  - BBD's European EPU has a statistically significant, albeit small, negative correlation with investment rates.
- DID (investment irreversibility as treatment intensity):
  - DID results (Columns (3)–(4)) show negative impact of policy uncertainty becomes larger, "around 9 to 13 percent," and are statistically significant for both quarterly EPU index choices.
  - Interpretation example: a one standard deviation increase in the quarterly EPU index around end of earlier year or beginning of current year induces firms in sectors with high investment irreversibility to reduce real annual investment rate by around 9 to 13 percentage points, holding other things constant.
  - Recent years: EPU index increased by about 0.2 to 0.5 per year, implying a decline in the real investment rate for firms in these sectors by about 2 to 6.5 percentage points.
  - Context: mean and median real investment is around 8 to 10 percent, respectively, for 2015.
- Alternative dependent variable (investment as share of total assets):
  - Baseline regression (Columns (5)–(6)): estimated impact of policy uncertainty on investment becomes smaller and is no longer statistically significant.
  - DID results (Columns (7)–(8)): negative DID impact remains statistically significant.
  - Coefficient estimates suggest a one standard deviation increase in the quarterly EPU index around end of previous year and beginning of current year could reduce investment-to-total-asset ratio in current year by around 6 percentage points.
- Additional timing evidence:
  - Table A.2 (Panels A and B) shows a weaker but statistically significant negative impact of policy uncertainty in the third quarter of the previous year on the real investment rate this year, suggesting investment decisions could be made as early as the second half of the earlier year.
- Selected coefficient estimates from Table 1 (preserved exactly as presented):
  - EPU index: -4.882**, -7.403, 0.424, -3.551, -1.978, -3.623, 0.874, -1.175 (with corresponding standard errors in parentheses).
  - EPU x Inv. irreversibility: -12.767**, -9.444*, -6.862**, -6.001* (with corresponding standard errors in parentheses).
  - L.Operating cashflows (% of capital stock): 0.130*, 0.164**, 0.130*, 0.167**, 0.087**, 0.103**, 0.086**, 0.105**.
  - Sales growth; real (%): 0.665***, 0.666***, 0.666***, 0.671***, 0.426***, 0.427***, 0.427***, 0.429***.
  - Borrowing cost; real (%): -1.328***, -1.611***, -1.356***, -1.655***, -0.677***, -0.829***, -0.692***, -0.857***.
  - BBD's European EPU: -0.151**, -0.057, -0.150**, -0.056, -0.067, -0.024, -0.067, -0.023.
  - One-year-ahead GDP forecast: 5.559, 7.655*, 5.509, 7.660*, 3.708, 4.640, 3.681, 4.644.
  - Observations: 383 (all specifications shown).
  - R2: 0.503, 0.505, 0.510, 0.508, 0.521, 0.524, 0.526, 0.527.
- Notes:
  - Other controls include Real GDP growth (%) and Inflation (%) and their one-period lags; and CDS spread.
  - Significance legend: * p<0.1; ** p<0.05; and *** p<0.01.
  - Clustered standard errors at the sector level in parentheses.
  - Regressions with time trends include linear and quadratic trends.

### Employment Decisions (Results and Interpretation)
- Table 2: employment growth used as dependent variable.
- Baseline results are in the opposite direction compared with previous literature.
- Including DID treatment intensity (EPU × investment irreversibility) makes estimated impact of policy uncertainty on employment growth negative, but DID specifications show no statistically significant effects.
- Interpretation caveats provided in the source:
  - Employment growth variable reflects aggregate number of employees only; it does not capture changes in composition of human capital stock or types of contracts.
  - With capital-skill complementarities, firms reducing investment in response to rising policy uncertainty may not recruit additional skilled workers.
  - Firms that postpone investment could substitute workers for investment in the short term, explaining potential negative correlation between hiring and uncertainty.
  - Turkey-specific labor supply dynamics: labour force grew at an average of 3.5 percent per annum during the past decade, about 4 percent per annum during 2010-2011, and above 6 percent in 2014 when policy uncertainty picked up—these structural changes in labour supply could affect relationship between policy uncertainty and employment growth.
  - Muted impact on employment may also be driven by structural changes towards greater reliance on capital accumulation after the Global Financial Crisis.
- Selected coefficient estimates from Table 2:
  - EPU index: 4.207, 8.358**, 5.595**, 9.570** (with standard errors in parentheses).
  - EPU x Inv. irreversibility: -3.341, -2.969 (with standard errors in parentheses).
  - Sales growth; real (%): 0.546*** (repeated across specifications).
  - Borrowing cost; real (%): -0.793***, -0.433, -0.800***, -0.447.
  - BBD's European EPU: -0.079, -0.177**, -0.079, -0.177**.
  - Observations: 383 (all specifications shown).
  - R2: 0.473, 0.483, 0.474, 0.484.

### Leverage Strategy (Net Borrowing Results)
- Baseline regressions (Columns (1) and (2)) show economic policy uncertainty is associated with higher net borrowing.
  - Possible interpretations: firms increase cash buffers and financing in uncertain times; or reverse causality if central bank policy uncertainty rises during periods of rapid growth/overheating.
- DID estimates (Columns (3) and (4)) show coefficient on uncertainty is negative and statistically significant.
  - A one standard deviation increase in the EPU index around the end of the previous year and the beginning of the current year would lead to around 17 to 21 percentage points reduction in the ratio of total net borrowing to total liabilities in the current year.
- Breakdown by maturity (Columns (5)–(8)):
  - Reduction in total net borrowing is driven by a reduction in net short-term borrowing.
  - Results for long-term borrowing are smaller and insignificant when using the EPU index of the last quarter of the preceding year.
  - Implication: economic policy uncertainty could lead to a change in the structure of firms’ liabilities.
- Timing: net borrowing of the current year is largely and negatively affected by policy uncertainty at the beginning of the year and at the end of the previous year; policy uncertainty two to three quarters earlier also negatively impacts net borrowing (Panels D, E, and F of Table A.2, Appendix).

### Robustness Checks
- DID relies on conditional common time trend: conditional on controls and absence of treatment, treatment group behaves like control group.
- Robustness checks conducted:
  - Alternative treatment intensity variables tested.
  - More restrictive time-dummy model applied to DID regressions.
  - Exclusion of sectors where public sector likely plays a role to re-examine impact for remaining sectors.
  - Use of contemporaneous and quarterly-lagged annualized EPU indices as regressors (instead of various quarterly EPU choices) to test whether economic decisions respond to current information only or also to recent history of policy uncertainty.
- Purpose: assess sensitivity of main findings to treatment definition, time controls, sector composition, and timing of EPU measures.

*Source: wp18272 - Section 3*

### Section 4

### wp18272 - Section 4

### Using Different Treatment Intensity Variables
- Two alternative DID treatment intensity variables are applied:
  - the ratio of machinery, plants, and equipment to total assets (narrower definition of investment irreversibility, excludes intangible fixed assets);
  - the depreciation rate, calculated as a ratio of total depreciation of tangible and intangible fixed assets and other long-term assets to total assets (interpreted as inverse treatment intensity because high depreciation implies lower upfront sunk costs).
- Main robustness findings:
  - Policy uncertainty induces lower investment and short-term borrowing among firms with high exposure to policy uncertainty, measured by the ratio of machinery, plants, and equipment to total assets (Columns (1) to (5) of Table A.3).
  - A one standard deviation increase in the EPU index at the beginning of the year would result in a 12-percentage point decline in the investment rate and around 20 percentage points lower net short-term borrowing (as a ratio of total short-term liabilities at the beginning of the year), compared to their counterfactual outcomes in absence of exposure to policy uncertainty.
  - Using the depreciation rate as the DID treatment intensity (i.e., identifying firms with low exposure) yields positive coefficient estimates (Columns (6) to (10) of Table A.3), confirming that firms relatively less affected by policy uncertainty cut less investment and borrowing when policy uncertainty increases.

### DID Regressions with Time Dummies
- Concern: regression estimates from over-time variation in the EPU index may capture other unobserved time-varying factors common to all sectors (e.g., domestic and external demand, economic prospects) despite inclusion of linear and quadratic trends.
- More restrictive specification: replace linear and quadratic time trends with time-dummies.
- Result: Table A.4 in the Appendix shows results are generally robust to this more restrictive specification.

### Exclusion of Basic Service and Construction Sectors
- Motivation: test whether negative impact of policy uncertainty is driven by sectors relatively more influenced by the public sector (fundamental service-related sectors such as health care, education, energy) and construction sector where Housing Development Administration of Turkey (TOKI) is active; these sectors may have a relatively higher share of irreversible investment and be structurally different.
- Result: Robust to exclusion of these sectors (Table A.5 in the Appendix).
  - A one standard deviation increase in the EPU index (at the end of the previous year and the beginning of the current year) induces a 9 to 13-percentage point decline in the real investment rate and a 20-percentage point reduction in total net borrowing as a share of total liabilities.

### Using the Annualized EPU Index as A Regressor
- Purpose: assess whether firms use collective information on policy uncertainty beyond specific quarters by using the annualized EPU index in panel DID regressions.
- Notation and lags:
  - U_{t−k}^A represents the annualized EPU index and k ∈ {0,1,2,3,4} stands for the quarter lag between the annualized EPU index and the outcome variable.
  - U_{t−0} is contemporaneous annualized EPU; U_{t−1} is one-quarter lagged annualized EPU (12-month counts start in October of year t−1 and end in September of year t); U_{t−2} starts in July of year t−1 and ends in June of year t; etc.
- Empirical findings (Table A.6 in the Appendix):
  - Statistically significant negative effects of the two- and three-quarter lagged annualized EPU index (Columns (3) and (4)) on real annual investment, total net borrowing, and net short-term borrowing.
  - Lower net long-term borrowing appears to be driven only by the three-quarter lagged annualized EPU index.
  - Interpretation: collective information on policy uncertainty from the third quarter last year onwards could influence investment and net borrowing in the current year.
- Caution: the contemporaneous annual impact (Column (1)) is likely subject to contemporaneous reverse causality; reverse causality weakens but remains present for longer-quarter lags of the annualized EPU index. This motivates use of quarterly EPU index in main specifications.

### Conclusion
- Uncertainty plays a key role in economic decision-making; news-based indices have established a strong and robust correlation between uncertainty and economic activity, but reverse causality and spurious correlation remain challenges.
- Turkey provides a useful case due to frequent periods of exogenous political-driven uncertainty and a reliable sectoral database; the authors construct a Turkey EPU index based on external news sources following Baker et al. (2016).
- Key empirical conclusions:
  - An increase in the first-quarter EPU index by one standard deviation is associated with a 5-percentage point lower real annual investment rate in the same year; the magnitude more than doubles for sectors with high degree of irreversibility.
  - Employment growth results indicate uncertainty may increase employment (firms may substitute labour for fixed capital in the short term); DID estimates show sectors with more irreversibility hire in net terms relatively less when uncertainty is high.
  - Sectors with high level of irreversibility relatively reduce short-term borrowings in periods of uncertainty.
- Policy implication: policy makers should strive to reduce policy uncertainty. The results call for extending similar sectoral analysis in other emerging markets where uncertainty remains high.

*Source: wp18272 - Section 4*

### Section 5

### wp18272 - Section 5

### References (selected)
- Rodrik, D., 1991, Policy Uncertainty and Private Investment in Developing Countries, Journal of Development Economics, Vol. 36 (2), p.229-242.
- Sahinoz, S. and E. E., Cosar, 2018a, Quantifying Uncertainty and Identifying Its Impacts on The Turkish Economy, Central Bank of the Republic of Turkey, CBRT Working Paper No. 18/06.
- Sahinoz, S. and E. E., Cosar, 2018b, Economic Policy Uncertainty and Economic Activity in Turkey, Applied Economics Letters, Vol. 25 (21), p.1517-1520.
- Samsar, A, 2011, Company Accounts Data of the Central Bank of the Republic of Turkey: Current Practices and Improvement Directions, IMF-OECD Conference on Strengthening Sectoral Position and Flow Data in the Macroeconomic Accounts.
- Turkish Statistical Institute, Main Statistics on National Accounts, Inflation and Price, and Employment, Unemployment, and Wages, Accessed on 18 May 2017, http://www.turkstat.gov.tr/UstMenu.do?metod=temelist
- Williams, J.C., 2013, The economy and fed policy: Follow the demand, Federal Reserve Bank of San Francisco, FRBSF Economic Letter 2013-05.
- Yasar, M., and C. J. M. Paul, 2008, Capital-Skill Complementarity, Productivity and Wages: Evidence from Plant-Level Data for A Developing Country, Labour Economics, Vol. 15, p.1-17.

### Appendix — Economic Policy Uncertainty Search Algorithms
- Database: Factiva’s Top Sources and all English-language Turkish news sources.
- Inclusion criteria (articles must satisfy all):
  - One of the three keyword mentions must be within 10 words of the word “Turkey” or “Turkish”.
  - Filtered by geographic coverage to cover only those tagged as being related to Turkey.
  - News articles must contain more than 99 words.
  - News articles must be tagged as related to economics and politics.
- Economic terms:
  - econom*
- Fiscal policy terms:
  - (government or treasury or MOF or ministry of finance or Hazine) or
  - ((spending or expenditure* or revenue*) near2 (government or public or fiscal or budget)) or
  - (austerity or tax* or automatic stabili?er) or
  - (debt near2 (public or national or sovereign or government)) or
  - (budget near2 (surplus or deficit* or plan* or balance* or gap)) or
  - deficit reduction or (fiscal near2 (policy or policies or plan* or cris?s or emergency or measure* or gap* or discipline* or consolidat* or stimulus or expansion*))
- Monetary and financial sector policy terms:
  - (monetary or monetary policy) or
  - (yield* or interest rate* or overnight rate* or discount rate* or interest*) or
  - (central bank or Central Bank of Turkey or Central Bank of the Republic of Turkey or TCMB or CBRT or Banking Regulation and Supervision Agency or BRSA) or
  - (regulat* or bank* supervis*) or
  - (quantitative easing or money supply or (tight* or loose*) near2 (money or monetary or policy)) or
  - (accommod* policy or monetary accommod*) or
  - (bond* purchas* or asset* purchas* or open market operation*)
- Uncertainty terms:
  - uncert* or (ambiguous or ambiguity) or dubious or precarious or unpredictable or undecided or undetermined or unresolved or unsettled or concern* or (worry* or worries) or (anxiety* or anxious) or doubt* or unclear
  - not (“without doubt” or “no uncertainty” or “no doubt” or reduced concern* or shares or equit* or stock market*)
- Normalizer:
  - Total counts of economics- and politics-related news articles in Factiva’s Top Sources and the English-language Turkish news sources, containing the common and neutral term “today”, and comprising more than 99 words.

### Robustness Checks — Summary of reported regression results and coefficients
- General notes across tables:
  - Note: 1/ * p<0.1; ** p<0.05; and *** p<0.01.
  - 2/ Clustered standard errors (at the sector level) in parentheses.
  - 3/ Regressions with time trends include linear and quadradic trends (where indicated).
  - 4/ All other controlling variables are similar to the main regressions in Tables 1 to 3 (where indicated).
  - FE indicates fixed effects; Time indicates Trend or Dummy per regression specification.
  - Observations reported: 383 in the panel regressions.

- Table A.1: Effects of Policy Uncertainty in Other Quarters of the Previous Year (selected coefficient reports)
  - Panel A. It/Kt-1 (%, In Real Terms) — examples:
    - EPU index: 0.424 (3.149)
    - EPU x Inv. irreversibility: -12.767** (5.168)
    - R2: 0.510
    - Linear Combination (EPU index + EPU x Inv. Irreversibility): -12.342*** (3.643)
  - Panel B. It/At-1 (%, In Real Terms) — examples:
    - EPU index: 0.874 (1.860)
    - EPU x Inv. irreversibility: -6.862** (3.021)
    - R2: 0.526
    - Linear Combination: -5.988** (2.335)
  - Panel C. Employment Growth (%) — examples:
    - EPU index: 5.595** (2.709)
    - EPU x Inv. irreversibility: -3.341 (3.312)
    - R2: 0.474
    - Linear Combination: 2.254 (3.346)
  - Quarterly EPU used in regression columns: Q1 t; Q4 t-1; Q3 t-1; Q2 t-1; Q1 t-1.
  - Time: Trend; FE: Yes; Observations: 383; Diff-in-Diff specification.

- Table A.2: Effects of Policy Uncertainty in Other Quarters of the Previous Year (panels D–F, selected)
  - Panel D. Ratio of Net Borrowing to Total Liabilities at t-1 (%, in USD) — examples:
    - EPU index: 14.078*** (4.119)
    - EPU x Inv. irreversibility: -21.064*** (6.735)
    - R2: 0.555
    - Linear Combination: -6.986* (3.783)
  - Panel E. Ratio of Net Short-Term Borrowing to Short-Term Liabilities at t-1 (%, in USD) — examples:
    - EPU index: 18.170*** (5.592)
    - EPU x Inv. irreversibility: -19.778** (7.931)
    - R2: 0.527
    - Linear Combination: -1.607 (5.484)
  - Panel F. Ratio of Net Long-Term Borrowing to Long-Term Liabilities at t-1 (%, in USD) — examples:
    - EPU index: 5.829 (4.153)
    - EPU x Inv. irreversibility: -19.029** (8.966)
    - R2: 0.365
    - Linear Combination: -13.200** (5.769)

- Table A.3: Alternative Treatment Intensity Variables (selected reported interactions and coefficients)
  - Interaction with MPE intensity:
    - L.EPU x MPE intensity: -12.096** (5.329); alternative: -6.419** (3.074); other reported: -2.514 (3.467); -21.323*** (6.688); -20.563** (7.891); -18.426** (8.951)
  - Interaction with Depreciation rate:
    - L.EPU x Depreciation rate: 16.575** (6.083); alternatives: 8.641** (3.452); 3.784 (3.941); 24.696*** (8.574); 24.655** (9.995); 20.913* (10.878)
  - R2 values reported across specifications range (examples): 0.509; 0.511; 0.525; 0.526; 0.474; 0.556; 0.554; 0.528; 0.364.
  - Quarterly EPU used in regression columns: Q1 t; Q4 t-1; Q3 t-1; Q2 t-1; Q1 t-1.
  - Time: Trend; FE: Yes; Observations: 383; Diff-in-Diff specification.
  - Panels correspond to outcomes: A. It/Kt-1 (%, In Real Terms); B. It/At-1 (%, In Real Terms); C. Employment Growth (%); D. Ratio of Net Borrowing to Total Liabilities at t-1 (% in USD); E. Ratio of Net Short-Term Borrowing to Short-Term Liabilities at t-1 (% in USD); F. Ratio of Net Long-Term Borrowing to Long-Term Liabilities at t-1 (% in USD).

- Table A.4: DID Regressions with Time Dummies (selected L.EPU x Inv. irreversibility coefficients)
  - Examples:
    - L.EPU x Inv. irreversibility: -12.676** (5.163); R2: 0.520
    - L.EPU x Inv. irreversibility: -6.808** (3.030); R2: 0.538
    - L.EPU x Inv. irreversibility: -3.353 (3.210); R2: 0.496
    - L.EPU x Inv. irreversibility: -20.934*** (6.654); R2: 0.576
    - L.EPU x Inv. irreversibility: -19.665** (7.745); R2: 0.559
    - L.EPU x Inv. irreversibility: -18.857** (9.000); R2: 0.373
  - Quarterly EPU used in regression: Q1 t; Q4 t-1; Q3 t-1; Q2 t-1; Q1 t-1.
  - Time: Dummy; FE: Yes; Observations: 383.
  - Panels correspond to outcomes: F, B, C, D, E, A (as labeled in the table rows).

- Table A.5: DID Regressions Excluding Some Sectors (examples; excluding construction, energy, and education, and health care sectors)
  - L.EPU x Inv. irreversibility coefficients (examples repeated from robustness specifications):
    - -12.676** (5.163); R2: 0.520
    - -6.808** (3.030); R2: 0.538
    - -3.353 (3.210); R2: 0.496
    - -20.934*** (6.654); R2: 0.576
    - -19.665** (7.745); R2: 0.559
    - -18.857** (9.000); R2: 0.373
  - Quarterly EPU used in regression: Q1 t; Q4 t-1; Q3 t-1; Q2 t-1; Q1 t-1.
  - Time: Trend; FE: Yes; Observations: 383.

- Table A.6: DID Regressions Using the Annualized EPU Index as A Regressor (selected L.EPU x Inv. irreversibility coefficients across Nq_Lagged_EPU = 0 to 4)
  - Nq_Lagged_EPU = 0 examples:
    - L.EPU x Inv. irreversibility: -1.862 (4.133); R2: 0.499
    - L.EPU x Inv. irreversibility: -2.657 (2.411); R2: 0.520
    - L.EPU x Inv. irreversibility: 3.520 (6.611); R2: 0.467
    - L.EPU x Inv. irreversibility: -6.870 (5.299); R2: 0.531
    - L.EPU x Inv. irreversibility: -6.028 (6.205); R2: 0.498
    - L.EPU x Inv. irreversibility: -5.509 (7.669); R2: 0.362
  - Nq_Lagged_EPU = 1 to 4 include statistically significant negative interactions in several specifications:
    - Example: Nq_Lagged_EPU = 2 — L.EPU x Inv. irreversibility: -9.400* (4.620); R2: 0.509
    - Example: Nq_Lagged_EPU = 3 — L.EPU x Inv. irreversibility: -10.554** (4.410); R2: 0.510
    - Example: Nq_Lagged_EPU = 1 — L.EPU x Inv. irreversibility: -5.524* (2.923); R2: 0.523
    - Example: Nq_Lagged_EPU = 3 — L.EPU x Inv. irreversibility: -18.013*** (5.956); R2: 0.561
    - Example: Nq_Lagged_EPU = 4 — L.EPU x Inv. irreversibility: -18.832** (8.015); R2: 0.546
  - Time: Trend; FE: Yes; Observations: 383.
  - Note 3/ Nq_Lagged_EPU reports the number of lagged quarters for the annualized EPU index.
  - Panels correspond to outcomes A–F as in previous tables.

### Key empirical patterns reported in the appendix tables (as presented)
- Several specifications report negative and statistically significant coefficients on the interaction term EPU x Inv. irreversibility (exact reported values include, among others, -12.767**, -21.064***, -19.778**, -20.934***, -19.665**, -18.857**, and others), with standard errors reported in parentheses for each coefficient.
- Several specifications report positive and statistically significant coefficients on the EPU index alone for borrowing-related outcomes (exact reported values include 14.078***, 18.170***, 18.822***, 27.388***, and others), with standard errors in parentheses.
- Alternative treatment intensity variables (MPE intensity and Depreciation rate) show large-magnitude interactions with L.EPU in multiple specifications; examples include L.EPU x MPE intensity: -21.323*** (6.688) and L.EPU x Depreciation rate: 24.696*** (8.574).
- Results are reported across multiple quarterly EPU lags (Q1 t; Q4 t-1; Q3 t-1; Q2 t-1; Q1 t-1), with robustness checks including time trends, time dummies, excluding sectors, and using an annualized EPU index with different numbers of lagged quarters (Nq_Lagged_EPU = 0 to 4).

*Source: wp18272 - Section 5 (appendix content and robustness tables).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18272.pdf_