## wpiea2021251-print-pdf

## Source details

**Canonical URL:** [wpiea2021251-print-pdf](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021251-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021251-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021251-print-pdf.pdf.json)

---

### Figures and Tables Inventory
- Figures:
  - 1. The GPRNK Index (1995-2020)
  - 2. The GPRNK Index by Topic
  - 3. Comparing the GPRNK Index with Existing Measure
  - 4. Assessing Potential Selection Bias of the GPRNK Index
- Tables:
  - 1. Search Keywords
  - 2. Variable Description
  - 3. Summary Statistics
  - 4. GPRNK and Firms’ Stock Return: Average Effects
  - 5. GPRNK and Firms’ Stock Return: Differential Effects by Firm Characteristics
  - 6. GPRNK and Firms’ Stock Return: Result with Subtopic Indices
  - 7. Regression Results with Alternative Measures
  - 8. Robustness Check: Results from Restricted Sample
- Appendices:
  - A.1. Details of Media Sample of GPRNK Index
  - A.2. Additional Robustness Checks
  - A.3. Additional Understanding of GPRNK Trend

### Major findings and motivation
- Rising geopolitical tensions (examples cited: U.S.-Iran conflicts over nuclear deals, terrorist attacks in Europe, Hong Kong-mainland China conflict, U.S. withdrawal from Afghanistan, Syria, Libya, Yemen) can increase financial market volatility and induce investment delays.
- South Korea context:
  - Market value of the Korean Stock Exchange amounted to USD $1.4 trillion at end-2019, the 15th biggest in the world and the sixth in Asia.
  - International investors own 37.2% of the KOSPI market and 13.3% of the KOSDAQ market.
  - North Korea’s unpredictability: six nuclear tests total, four since 2012.
- Core empirical finding:
  - Firm-level panel regressions with firm fixed effects show heightened geopolitical risk (GPRNK) reduces a company’s stock returns, with effects robust to controls for market volatility and Economic Policy Uncertainty (EPU).
- Heterogeneous effects:
  - Firms with a large share of fixed assets: stock returns depressed when GPRNK increases.
  - Firms with experience in inter-Korean economic cooperation: stock returns depressed when GPRNK increases.
  - Firms with a large share of international investors: relatively higher stock returns during tension-increasing periods.
- Interpreted channels: international portfolio diversification and investment irreversibility.

### Construction of the GPRNK index — data and transformation
- Media sample:
  - BigKinds database (approximately 60 million articles across 54 media outlets); selected 18 representative South Korean media outlets (10 national daily newspapers, 5 business/economics newspapers, 3 national broadcasters); topics limited to politics, economics, and international relations.
- Keyword selection (five-step process):
  1. Default keyword: "North Korea".
  2. Four topic categories: military tensions, sanctions, talks/agreements, economic cooperation.
  3. Human reading around major events to list topic and action/status keywords.
  4. Exclude words that collocate but negate original aspects to avoid false positives.
  5. Iterative validation to finalize recurring keywords.
- Search keyword counts (Table 1):
  - Default keywords ("North Korea"): 1,039,297 articles.
  - Military tensions (negative): 124,328 articles.
  - Sanctions (negative): 24,637 articles.
  - Talks/agreements (positive): 124,824 articles.
  - Economic cooperation (positive): 16,774 articles.
- Monthly media-specific measure (notation from source):
  - N_{j,i,t} = total number of articles in category j from media i at time t.
  - N_neg,i,t and N_pos,i,t combine negative and positive categories.
  - N_{i,t} = total number of articles with default keyword "North Korea" for media i and time t.
  - X_{i,t} = (N_neg,i,t − N_pos,i,t) / N_{i,t}
  - X_hat_{i,t} = 1/2 * (X_{i,t} + sqrt(X_{i,t}^2 + 0.1))
  - Y_{i,t} = X_hat_{i,t} / sigma_i, where sigma_i is the time-series within-newspaper standard deviation of X_hat_{i,t} from January 1995 to December (text truncates).
- Aggregation and normalization:
  - For each month, Y_i = 1/N Σ_{i=1}^N Y_{i,i,t} where N is the total number of media outlets.
  - GPRNK index normalized to mean of 100: GPRNK_t = 100 * Y_t / Y_bar where Y_bar is the mean Y_t from 1995 to 2016.
- Transformation choice:
  - f(x) = 1/2 (x + sqrt(x^2 + α)), α = 0.1 (monotonic, convex, asymptotic properties noted).

### Evaluating the GPRNK index (1995–2020)
- Temporal behavior:
  - Index plotted January 1995 to November 2020.
  - Sharp increases during nuclear tests, missile launches, or military confrontations; decreases around bilateral or multilateral meetings.
  - Largest spikes during the North Korea nuclear/missile crisis of 2017.
- Category composition:
  - Negative and positive indices dominated by military tension and talks categories.
  - Correlations:
    - negative GPRNK vs military tension = 0.99.
    - positive GPRNK vs talks = 0.98.
    - correlation between positive-domain indices: 0.43.
    - overall GPRNK correlation with Korea Peace Index (KOPI) ≈ 0.57.
    - GPRNK correlation with GPR Korea index (Caldara and Iacoviello, 2019) ≈ 0.48.
- Comparisons with other indices:
  - KOPI tracks actual geopolitical events and may underrate threats from WMD development.
  - GPR Korea (English-language, general keywords) may miss reductions in risk during summit meetings and gradual escalation in 2016–2017.

### Robustness and potential sample biases
- Media coverage:
  - Media sample covers 68.9% of total newspaper circulation and 63.6% of total viewership among national broadcast channels.
  - Three high-circulation, right-leaning newspapers have shorter sample periods: Chosun Ilbo (2018-), Dong-a Ilbo (2018-), Joong-ang Ilbo (2008-).
  - Excluding those three yields correlation of 0.98 with the benchmark index (no significant selection bias).
  - Right-leaning vs left-leaning newspapers co-move closely: correlation coefficient = 0.94.

### Empirical framework to estimate impact on stock returns
- Baseline monthly panel regression (firm fixed effects and aggregate controls):
  - Return_{i,t} = α_i + β · GPRNK_t + X_{t}' γ + ε_{i,t}
    - Controls X_t: monthly returns of the Dow Jones Industrial Average, nominal exchange rate against the U.S. dollar, dummy for Korean financial crisis (1997-1998), dummy for Global Financial Crisis (2008-2009).
  - Additional controls considered: VKOSPI (implied volatility index of Korean stock market) and Korean EPU.
- Heterogeneity specification (interactions to identify channels):
  - Return_{i,t} = α_i + F_{i,t-1}' β1 + GPRNK_t · F_{i,t-1}' β2 + C_{i,t}' β3 + δ_t + ε_{i,t}
    - F_{i,t-1}: lagged firm characteristics.
    - C_{i,t}: controls (leverage ratio, book-to-market ratio, profitability).
    - Firm characteristics examined:
      - Fixed assets / total assets (investment irreversibility).
      - Foreigner ownership.
      - Firm size (log assets).
      - Economic cooperation dummy.
      - Defense industry dummy.

### Data, sample selection, and key summary statistics
- Sample:
  - 3,251 South Korean non-financial companies between 1996 to 2019 initially.
  - Stock market data: DataGuide (Fnguide).
  - Accounts data: KIS Value (NICE).
- Screening:
  - Exclude firms observed < 10% of sample period.
  - Drop firms with any capital impairment.
  - Eliminate observations with operational losses exceeding total assets.
  - Final sample: 2,467 firms.
- Selected variable definitions (Table 2):
  - Stock Returns: Log difference in last day’s revised stock price between period t and period t-1 (DataGuide).
  - Price-to-Book Ratio (PBR): Ratio of share price to book value per share at end of month.
  - Foreigner Ownership: Proportion of foreigners’ share in market capital.
  - Asset Size: Log of total value of asset at end of quarter.
  - Return on Assets (ROA): Ratio of operating profit to total asset value.
  - Fixed-to-Total Assets: Ratio of fixed capital to total asset value.
  - Leverage Ratio: Ratio of total debt to market capitalization.
  - Economic Cooperation: Dummy for involvement in economic cooperation (Kim and Jung (2014)).
  - Defense: Dummy for defense-related business (Kim and Jung (2014)).
- Summary statistics (Table 3 narrative):
  - Sample average monthly stock returns over whole sample period: -0.7%.
  - Average monthly stock returns when GPRNK index > 100: -0.9%.
  - Average monthly stock returns when GPRNK index < 100: -0.4%.

### Results — baseline and controls (Section 4.1)
- Baseline regression (1996m2~2019m12; obs: 400,818; firms: 2,467; Adjusted R2: 0.0397):
  - Log(GPRNK): -0.0068*** (0.0007)
  - Economic interpretation: A one log point increase in geopolitical risk reduces stock returns by 0.0068%p on average.
  - Example: A decrease in the GPRNK index of 123 log points (magnitude observed 2017→2018, ≈ three standard deviations) associated with an increase in monthly stock returns of 0.83%p on average.
- Controlling for Log(VKOSPI) (2003m1~2019m12; obs: 329,356; firms: 2,402; Adjusted R2: 0.0337):
  - Log(GPRNK): -0.0158*** (0.0009)
  - Log(VKOSPI): -0.0158*** (0.0012)
- Controlling for Log(EPU):
  - Log(GPRNK): -0.0019** (0.0007)
  - Log(EPU): -0.0225*** (0.0007)
- Pooling Log(GPRNK), Log(VKOSPI), Log(EPU): all three remain statistically significant (Column (4), Adjusted R2: 0.0352; obs: 329,356; firms: 2,402).
- Adding global GPR and global EPU: similar results; Log(GPRNK): -0.0148*** (0.0009) (Column (5); obs: 329,356; firms: 2,402; Adjusted R2: 0.0374).

### Differential effects by firm characteristics (Table 5)
- Interaction effects (coefficients are effects of 1 log-point increase of GPRNK; foreigner ownership, log(asset), fixed-to-total asset ratio standardized to unit standard deviation):
  - Log(GPRNK) * foreigner ownership = 0.0034*** (0.0009)
    - Interpretation: A firm with one-standard-deviation-larger foreign ownership experiences 0.0034%p higher returns per one log-point increase in GPRNK.
  - Log(GPRNK) * log(asset) = -0.0005 (0.0008) (insignificant)
  - Log(GPRNK) * fixed asset/total asset = -0.0017** (0.0007)
    - Interpretation: One-standard-deviation-greater fixed-asset share increases negative effect by 0.0017%p.
  - Log(GPRNK) * economic cooperation dummy = -0.0142*** (0.0028)
    - Interpretation: Firms involved in economic cooperation see stock returns reduced by 0.0142%p more per one log point increase in GPRNK.
  - Log(GPRNK) * defense dummy = -0.0039 (0.0035) (insignificant)
- Pooled interactions (all included) amplify some coefficients (foreign ownership roughly doubles).
- Aggregate magnitudes for a two-standard-deviation increase in GPRNK:
  - Increase of 0.49%p for a firm with one-standard-deviation-higher foreign ownership.
  - Decrease of 0.19%p for a firm with one-standard-deviation-higher asset size.
  - Decrease of 0.16%p for a firm with one-standard-deviation-higher fixed-asset share.
  - Decrease of 1.16%p for a firm involved in economic cooperation with North Korea.

### Subtopic indices and asymmetric effects (Table 6)
- Negative-news index interactions:
  - Log(Index) * foreigner ownership = 0.0020*** (0.0006)
- Positive-news index interactions:
  - Log(Index) * foreigner ownership = -0.0054*** (0.0009)
  - Signs of interactions reverse relative to negative-news index; absolute values larger for positive-news index (asymmetric effects).
- Category-specific findings (Military threat, Sanctions, Talks, Economic Cooperation):
  - Firms with large foreign ownership:
    - Positive returns to military threats or sanctions news.
    - Negative returns to dialogue or economic cooperation news.
  - Asset-size effects generally insignificant except for “talk” category.
  - Fixed-to-total assets ratio:
    - Positive for “good news” categories; negative for “sanctions” category.
  - Firms involved in business with North Korea: sensitive to most category-specific measures except “sanctions”.
- Sample for subtopic regressions: 1996m2~2019m12; obs: 391,405; firms: 2,426; Adjusted R2 ~ 0.1695–0.1696.

### Alternative measures of geopolitical risk (Table 7)
- Panel A — Average effects (obs: 400,818; firms: 2,467):
  - Benchmark GPRNK: Log(GPRNK) = -0.0068*** (0.0007)
  - NK-to-All Index (frequency of North Korea-related articles): Log(Index) = -0.0207*** (0.0008)
  - GPR Korea Index (Caldara and Iacoviello, 2019): Log(Index) = -0.0165*** (0.0008)
- Panel B — Differential effects:
  - NK-to-All Index yields interaction signs opposite to baseline for asset size, fixed asset ratio, and economic cooperation, suggesting inability to separate upside vs downside news.
  - GPR Korea Index: Log(Index) * foreigner ownership = 0.0035*** (0.0010) (similar to benchmark), other interactions not significant.
- Interpretation: Benchmark GPRNK (net negative index using positive and negative news) better captures directionality than simple frequency measures.

### Robustness checks (Table 8 and appendices)
- Restricted-sample checks (all include firm and time fixed effects; controls include PBR, leverage, ROA; first-order terms of firm-level sensitivities included):
  - Exclude delisted firms as of Dec. 2019 (obs: 331,168; firms: 1,836; Adjusted R2: 0.2074)
    - Log(GPRNK) * foreigner ownership = 0.0052*** (0.0009)
  - Manufacturing firms only (obs: 256,464; firms: 1,598; Adjusted R2: 0.1752)
    - Log(GPRNK) * foreigner ownership = 0.0044*** (0.0012)
  - Exclude January and December (obs: 326,679; firms: 2,426; Adjusted R2: 0.1603)
    - Log(GPRNK) * foreigner ownership = 0.0057*** (0.0011)
  - Exclude extreme dependent values (-100 to 100 log-point) (obs: 390,336; firms: 2,426; Adjusted R2: 0.2056)
    - Log(GPRNK) * foreigner ownership = 0.0054*** (0.0009)
  - Market Return > 0 (obs: 216,128; firms: 2,426; Adjusted R2: 0.1036)
    - Log(GPRNK) * foreigner ownership = 0.0063*** (0.0014)
  - Market Return < 0 (obs: 175,277; firms: 2,426; Adjusted R2: 0.1574)
    - Log(GPRNK) * foreigner ownership = 0.0062*** (0.0016)
  - Sample Year >= 2003 (obs: 325,513; firms: 2,379; Adjusted R2: 0.1357)
    - Log(GPRNK) * foreigner ownership = 0.0052*** (0.0011)
    - Fixed-asset share coefficient becomes smaller and insignificant.
  - Sample Year >= 2012 (obs: 165,060; firms: 2,012; Adjusted R2: 0.1248)
    - Log(GPRNK) * foreigner ownership = 0.0046*** (0.0017)
    - Only foreign ownership and economic cooperation dummy interactions remain significant.
- Additional robustness:
  - Media-selection weighting, alternative windowing (three-month moving averages: windows [t-1,t+1], [t-2,t], weighting options), subscription-weighted index, standardized index yield similar results (Table A-5 and related appendices).
  - Controls for VKOSPI interactions and EPU interactions do not materially change main GPRNK interaction effects (Tables A-3, A-4).

### Conclusions and policy-relevant insights (Section 5)
- Developed a country-specific geopolitical risk measure focused on inter-Korean relations (GPRNK) capturing both tension-increasing and tension-decreasing moments.
- Dynamics:
  - Peak geopolitical tension in 2017 nuclear/missile crisis.
  - Index rises sharply with nuclear tests, missile launches, military aggression; drops around bilateral/multilateral talks and the first inter-Korean summit.
- Firm-level implications:
  - Heightened GPRNK depresses South Korean stock returns overall.
  - Negative effects particularly pronounced for:
    - firms with larger share of domestic investors,
    - firms with higher fixed assets/total assets,
    - firms involved in inter-Korean economic cooperation.
  - Evidence consistent with investment irreversibility and international portfolio diversification channels.
- Contribution:
  - GPRNK provides a Korea-specific geopolitical risk measure that improves on simple frequency indices by distinguishing negative vs positive news and by capturing inter-Korean dynamics relevant for firm valuations and policy assessment.

*Source: wpiea2021251-print-pdf*

### REFERENCES ___________________________________________________________________________________ 27

### REFERENCES

### Figures and Tables Inventory
- Figures:
  - 1. The GPRNK Index (1995-2020)
  - 2. The GPRNK Index by Topic
  - 3. Comparing the GPRNK Index with Existing Measure
  - 4. Assessing Potential Selection Bias of the GPRNK Index
- Tables:
  - 1. Search Keywords
  - 2. Variable Description
  - 3. Summary Statistics
  - 4. GPRNK and Firms’ Stock Return: Average Effects
  - 5. GPRNK and Firms’ Stock Return: Differential Effects by Firm Characteristics
  - 6. GPRNK and Firms’ Stock Return: Result with Subtopic Indices
  - 7. Regression Results with Alternative Measures
  - 8. Robustness Check: Results from Restricted Sample
- Appendices:
  - A.1. Details of Media Sample of GPRNK Index
  - A.2. Additional Robustness Checks
  - A.3. Additional Understanding of GPRNK Trend

### Major Findings (from Introduction and Summary of Methods)
- Motivation and context:
  - Rising geopolitical tensions (examples cited: U.S.-Iran conflicts over nuclear deals, terrorist attacks in Europe, Hong Kong-mainland China conflict, U.S. withdrawal from Afghanistan, Syria, Libya, Yemen) can increase financial market volatility and induce investment delays.
  - South Korea provides a favorable setting to study geopolitical risk because:
    - Market value of the Korean Stock Exchange amounted to USD $1.4 trillion at end-2019, the 15th biggest in the world and the sixth in Asia.
    - Nearly 40% of the market is owned by international investors (international investors own 37.2% of the KOSPI market and 13.3% of the KOSDAQ market).
  - South Korea has been exposed to high geopolitical risk due to North Korea’s unpredictable threats; six nuclear tests total, four since 2012.
- Construction of GPRNK index:
  - Novel index measuring geopolitical risk originating from North Korean threats.
  - Based on automated keyword searches from news articles in leading Korean-language newspapers and broadcasts since 1995.
  - Four main drivers: military conflicts, sanctions, talks, and economic engagement.
  - Index captures both downside and upside geopolitical risk by computing the relative frequency of net negative news articles compared to total North Korea–related articles.
- Empirical approach and main results:
  - Firm-level panel regressions of firm stock returns on GPRNK with firm fixed effects show that heightened geopolitical risk reduces a company’s stock returns.
  - Adverse effects remain significant after controlling for overall market volatility and economic policy uncertainty.
  - Heterogeneous effects:
    - Increased geopolitical risk depresses stock returns for companies with a large share of fixed assets.
    - Increased geopolitical risk depresses stock returns for companies with experience in inter-Korean economic cooperation.
    - Companies with a large share of international investors exhibit relatively high stock returns during tension-increasing periods.
  - Interpreted channels: international portfolio diversification and investment irreversibility.

### Relation to Existing Literature and Advantages
- Builds on text-based uncertainty measures:
  - Baker, Bloom and Davis (2016) (EPU index) and Caldara and Iacoviello (2019) (GPR index).
- Distinctions and contributions:
  - Focus on a region-specific, Korea-centric geopolitical risk measure rather than global GPR.
  - Uses Korean-language media to capture Korea-specific context (North Korea WMD development).
  - Captures both tension-escalating (negative) and peace-seeking (positive) events, unlike many prior indices that reflect primarily downside risk.
  - Addresses limitations of event studies (which focus on discrete events) by providing continuous measure of geopolitical intensity and by covering negotiation stages and verbal exchanges that may be priced in before events.

### Empirical Implications and Policy-Relevant Insights
- Key economic channels identified:
  - Investment irreversibility: political instability can cause firms to delay capital investment, weakening growth outlook and stock prices.
  - Investor behavior: international diversification can lead foreign investors to outperform domestic investors in some geopolitical shocks; information asymmetry can lead the reverse in other contexts.
- Policy implications for South Korean financial markets:
  - High openness to global investors (37.2% KOSPI, 13.3% KOSDAQ foreign ownership) implies geopolitical swings transmit internationally.
  - Understanding heterogeneity across firms (fixed assets share, prior inter-Korean cooperation exposure, foreign investor share) is important for assessing vulnerability to geopolitical shocks.

*Source: wpiea2021251-print-pdf - REFERENCES*

### 2016. The above standardized index can be averaged across media outlets.

### wpiea2021251-print-pdf - 2016. The above standardized index can be averaged across media outlets.

### Construction of the GPRNK index
- For each month, aggregate media-specific standardized measures:
  - 푌푌푖 = 1 푁푁 �푌푌푖푖푖푖 푁푁 푖푖=1 where 푁푁 is the total number of media outlets.
- Normalize to obtain the geopolitical risk index associated with the inter-Korea relationship with a mean of 100:
  - 퐺퐺퐺퐺퐺퐺 푁푁퐺퐺푖 = 100푌푌푖 푌푌� where 푌푌� is the mean value of 푌푌푖 from 1995 to 2016.
- Transformation choice for related series:
  - f(x) = 1 2 (x+ √(x2+α)) , (α>0) is a monotonic and convex transformation asymptotic to y=x as x → +∞ and to y=0 as x → -∞.
  - Parameter choice: α = 0.1 to ensure the new series 푋푋횤횤푖푖� has a similar distribution with the relative frequency variables and enough variation for 푋푋푖푖푖푖 < 0.

### Evaluating the GPRNK index (1995–2020)
- Temporal behavior:
  - The index is plotted from January 1995 to November 2020.
  - Sharp increases during occurrences of nuclear tests, missile launches, or military confrontations.
  - Significant decreases around bilateral or multilateral meetings.
  - Largest spikes observed during the North Korea nuclear/missile crisis of 2017.
- Category composition:
  - Negative and positive indices are mostly dominated by the military tension and talks categories.
  - Correlation coefficients:
    - Between negative-domain indices: 0.69 (overall negative vs positive domains stated earlier), detailed pair correlations: negative GPRNK vs military tension = 0.99; positive GPRNK vs talks = 0.98.
    - Between positive-domain indices: 0.43.
- Comparisons with other indices:
  - Korea Peace Index (KOPI):
    - Constructed from Yonhap News Agency articles, daily scores and monthly averages (January 2005 to September 2012).
    - Spikes at incidents like the Cheonan sinking and Yeonpyeong Island attack; plummets during six-party talks and inter-Korean summits.
    - Underrates tensions from WMD development because it weights actual geopolitical events more than threats.
    - Correlation with GPRNK: about 0.57.
  - GPR Korea index (Caldara and Iacoviello, 2019):
    - Uses automated text-searches with general keywords applicable worldwide; sourced from 11 leading English-language newspapers in the U.S., U.K., and Canada.
    - Peaks around North Korea’s withdrawal from the Non-Proliferation Treaty and threats to nullify the 1953 armistice — events more significant in global diplomacy than in inter-Korea context.
    - Fails to capture reductions in risk during summit meetings and the gradual escalation in 2016 and 2017.
    - Correlation with GPRNK: about 0.48.

### Robustness and potential sample biases
- Media coverage and sample notes:
  - Media sample covers 68.9% of total newspaper circulation and 63.6% of total viewership among national broadcast channels.
  - Three high-circulation, right-leaning newspapers have shorter sample periods in the data source: Chosun Ilbo (2018-), Dong-a Ilbo (2018-), Joong-ang Ilbo (2008-).
- Sensitivity checks:
  - Excluding the three newspapers with short time series yields a correlation of 0.98 with the benchmark index — exclusion does not create significant selection bias.
  - Subgroup indices by political leaning:
    - Right-leaning vs left-leaning newspapers co-move closely with a correlation coefficient of 0.94.
  - Note: Right-leaning newspapers list: Chosun Ilbo, Dong-a Ilbo, Joong-ang Ilbo, Kookmin Ilbo, Munhwa Ilbo. Left-leaning: Kyunghyang Sinmun, Hankyoreh.

### Empirical framework to estimate impact on stock returns
- Baseline monthly panel regression (firm fixed effects and aggregate controls):
  - 퐺퐺푖푖푖푖 = 훼훼푖 + 훽훽∙퐺퐺퐺퐺퐺퐺 푁푁퐺퐺푖 + 푋푋푖′ 훾훾 + 휀휀푖푖푖푖
    - 퐺퐺푖푖푖푖: monthly stock return of firm i at time t.
    - 퐺퐺퐺퐺퐺퐺 푁푁퐺퐺푖: GPRNK index.
    - 푋푋푖: time-varying aggregate controls — monthly returns of the Dow Jones Industrial Average Index, nominal exchange rates against the U.S. dollar, dummy for the Korean financial crisis in 1997-1998, dummy for the Global Financial Crisis in 2008-2009.
    - 훼훼푖: firm fixed effects.
  - Additional controls considered: VKOSPI (implied volatility index of the Korean stock market) and the Korean Economic Policy Uncertainty index (EPU).
- Heterogeneity specification to identify channels:
  - 퐺퐺푖푖푖푖 = 훼훼푖 + 퐹퐹푖,푖−1′ 훽1 + 퐺퐺퐺퐺퐺퐺 푁푁퐺퐺푖 ∙ 퐹퐹푖,푖−1′ 훽2 + 퐶퐶푖푖푖푖′ β3 + 훿훿푖 + 휀휀푖푖푖푖
    - 퐹퐹푖,푖−1: vector of firm characteristics lagged by one month.
    - 퐶퐶푖푖푖푖: control variables (leverage ratio, book-to-market ratio, profitability).
    - 훼훼푖 and 훿훿푖: firm and time fixed effects.
  - Firm-level characteristics examined (interaction with GPRNK):
    - Degree of investment irreversibility (fixed assets to total assets ratio).
    - Foreigners’ share of stock investment.
    - Firm size.
    - Involvement in economic cooperation with North Korea (dummy).
    - Whether firm is in the defense industry (dummy).
- Motivation and hypotheses behind firm-level channels:
  - Investment irreversibility:
    - Firms with a large fraction of fixed assets may delay investment under high uncertainty, lowering growth outlook and stock prices.
    - Caveat noted: fixed assets fraction may not always proxy irreversibility if assets are redeployable.
  - Foreign ownership:
    - Foreign participation in 2017: 33.6% of total market capitalization.
    - Hypotheses:
      - If information disadvantage dominates, foreign investors may be net sellers during high geopolitical risk → firms with large foreign ownership see lower returns.
      - If international diversification motives dominate, foreign investors may absorb risk → firms with higher foreign ownership see higher returns.
    - Distribution: at end of 2017, about half of 2,313 KRX-listed stocks have foreign ownership < 2%; 5% have more than 32%.
  - Firm size:
    - Mixed prior evidence: smaller firms more sensitive due to lending constraints; large firms possibly more negatively affected due to insufficient internal funding.
  - Economic cooperation and defense sector:
    - Firms involved in inter-Korean economic cooperation (e.g., those linked to Kaesong Industrial Complex) expected to have lower returns when GPRNK increases.
    - Defense firms might see increased returns when GPRNK increases.

### Data, sample selection, and key summary statistics
- Sample coverage and sources:
  - 3,251 South Korean non-financial companies between 1996 to 2019 in original sample.
  - Stock market data: DataGuide provided by Fnguide.
  - Accounts data: KIS Value provided by National Information & Credit Evaluation (NICE).
- Screening procedure to reduce influence of extreme observations:
  - Exclude firms observed during less than 10% of the sample period.
  - Drop firms that experienced any capital impairment.
  - Eliminate observations with operational losses exceeding total assets.
  - Final sample after screening: 2,467 firms.
- Variable definitions (selected from Table 2):
  - Stock Returns: Log difference in last day’s revised stock price between period t and period t-1 (DataGuide).
  - Price-to-Book Ratio (PBR): Ratio of share price to book value per share at end of month (DataGuide, KisValue).
  - Foreigner Ownership: Proportion of foreigners’ share in market capital (DataGuide).
  - Asset Size: Log of total value of asset at end of quarter (KisValue).
  - Return on Assets (ROA): Ratio of operating profit to total asset value (KisValue).
  - Fixed-to-Total Assets: Ratio of fixed capital to total asset value (KisValue).
  - Leverage Ratio: Ratio of total debt to market capitalization (KisValue).
  - Economic Cooperation: Dummy for involvement in economic cooperation (Kim and Jung (2014)).
  - Defense: Dummy for business related to defense industry (Kim and Jung (2014)).
- Summary statistics highlights (from Table 3 narrative):
  - Sample average of monthly stock returns over whole sample period: -0.7%.
  - Average monthly stock returns when GPRNK index > 100: -0.9%.
  - Average monthly stock returns when GPRNK index < 100: -0.4%.
  - Note: These unconditional averages do not imply causation; subsequent regression analysis with controls is used to estimate the average effect of geopolitical risk on stock returns.

*Italic: Content extracted from wpiea2021251-print-pdf - 2016. The above standardized index can be averaged across media outlets.*

### 4.   Results

### 4.   Results

### 4.1. Baseline Results
- Regressing firm-level monthly stock returns on Log(GPRNK) with time-varying controls and firm fixed effects:
  - Log(GPRNK): -0.0068*** (0.0007)
  - Sample: 1996m2~2019m12 for Column (1)
  - Number of observations: 400,818
  - Number of firms: 2,467
  - Adjusted R2: 0.0397
- Economic interpretation:
  - A one log point increase in geopolitical risk reduces stock returns by 0.0068%p on average.
  - A decrease in the GPRNK index of 123 log points (the magnitude observed from 2017 to 2018, approximately equivalent to three standard deviations) is associated with an increase in monthly stock returns of 0.83%p on average, controlling for other factors.
- Controlling for market implied volatility (Log(VKOSPI)):
  - Log(GPRNK) coefficient increases in absolute value: -0.0158*** (0.0009)
  - Log(VKOSPI): -0.0158*** (0.0012)
  - Sample period for regressions with VKOSPI: 2003m1~2019m12
  - Number of observations: 329,356
  - Number of firms: 2,402
  - Adjusted R2: 0.0337
  - Interpretation: GPRNK likely negatively correlated with VKOSPI; VKOSPI may not capture short-lived geopolitical shocks.
- Controlling for economic policy uncertainty (Log(EPU)):
  - Log(GPRNK): -0.0019** (0.0007)
  - Log(EPU): -0.0225*** (0.0007)
  - Interpretation: Correlation between GPRNK and EPU is positive; EPU effects larger than GPRNK effects.
- Pooling Log(GPRNK), Log(VKOSPI), Log(EPU):
  - All three uncertainty measures remain statistically significant.
  - Column (4) adjusted R2: 0.0352 (Number of observations: 329,356; Number of firms: 2,402)
- Adding global GPR and global EPU (Column (5)):
  - Results align with Column (4) except Log(EPU) (Korea) becomes insignificant.
  - Log(GPRNK): -0.0148*** (0.0009)
  - Number of observations: 329,356
  - Number of firms: 2,402
  - Adjusted R2: 0.0374
- Table 4 notes:
  - Dependent variable: monthly stock return.
  - Figures are fixed effect OLS coefficients; parentheses are robust standard errors clustered at the firm level.
  - *, **, *** indicate statistical significance at the 10%, 5%, 1% level, respectively.
  - Every coefficient represents the effect of 1 log-point increase of the uncertainty measure on stock returns.
  - Controls include monthly return in Dow Index, Won-Dollar exchange rate, global financial crisis and the Korean foreign currency crisis dummy variables.

### Differential Effects by Firm Characteristics (Table 5)
- Time fixed effects included; Log(GPRNK) collinear with time dummies and drops out in these specifications.
- Interaction results (coefficients are interaction effects of 1 log-point increase of GPRNK; Foreigner ownership, log(asset) and fixed-to-total asset ratio standardized to unit standard deviation):
  - Column (1): Log(GPRNK) * foreigner ownership = 0.0034*** (0.0009)
    - Interpretation: A firm with a one-standard-deviation-larger share of foreign ownership experiences 0.0034%p higher stock returns when GPRNK increases by one log point.
  - Column (2): Log(GPRNK) * log(asset) = -0.0005 (0.0008) (insignificant)
  - Column (3): Log(GPRNK) * fixed asset/total asset = -0.0017** (0.0007)
    - Interpretation: Firms with a larger share of fixed assets suffer more negative effects; a one-standard-deviation-greater share increases negative effect by 0.0017%p.
  - Column (4): Log(GPRNK) * ecoop dummy = -0.0142*** (0.0028)
    - Interpretation: Firms involved in economic cooperation with North Korea see stock returns reduced by 0.0142%p more per one log point increase in GPRNK relative to non-involved firms.
  - Column (5): Log(GPRNK) * defense dummy = -0.0039 (0.0035) (insignificant)
  - Column (6) (all interactions pooled):
    - Interaction signs and significance similar; Log(GPRNK) * log(asset) becomes significant and about four times larger in absolute value compared to separate regressions.
    - Log(GPRNK) * foreigner ownership roughly doubles in magnitude relative to separate regression.
- Aggregate interpretive magnitudes:
  - An increase in the GPRNK by two standard deviations is associated with:
    - Increase in stock returns of 0.49%p for a firm with a one-standard-deviation-larger share of foreigners.
    - Decrease of 0.19%p for a firm with a one-standard-deviation-larger amount of assets.
    - Decrease of 0.16%p for a firm with a one-standard-deviation-larger share of fixed assets.
    - Decrease of 1.16%p for a firm involved in economic cooperation with North Korea.

### Subtopic Indices (Table 6) — Category-specific GPRNK measures
- Column (1): Negative news index:
  - Log(Index) * foreigner ownership = 0.0020*** (0.0006)
  - Interaction terms broadly similar to benchmark, but Log(Index) * log(asset) smaller and statistically insignificant.
- Column (2): Positive news index:
  - Log(Index) * foreigner ownership = -0.0054*** (0.0009)
  - Signs of interaction coefficients reverse relative to benchmark for positive-news index.
  - Absolute values of estimated coefficients larger in regression of positive GPRNK index, implying asymmetric effects.
- Columns (3)–(6): Category-specific indices (Military threat, Sanctions, Talks, Economic Cooperation):
  - Firms with large foreign ownership:
    - Positive returns to military threats or sanctions news.
    - Negative returns to dialogue or economic cooperation news.
  - Asset size effects:
    - Insignificant except for “talk” category.
  - Fixed-to-total assets ratio:
    - Positive for “good news” categories; negative for “sanctions” category.
  - Firms involved in business with North Korea:
    - Sensitive to category-specific measures except for news on “sanctions”.
- Sample: 1996m2~2019m12; Number of observations: 391,405; Number of firms: 2,426; Adjusted R2 ~ 0.1695–0.1696.

### Alternative Measures of Geopolitical Risk (Table 7)
- Panel A — Average Effects:
  - Benchmark GPRNK: Log(GPRNK) = -0.0068*** (0.0007)
  - NK-to-All Index (relative frequency of North Korea-related articles): Log(Index) = -0.0207*** (0.0008)
  - GPR Korea Index (Caldara and Iacoviello, 2019): Log(Index) = -0.0165*** (0.0008)
  - All regressions include firm fixed effects; Number of observations: 400,818; Number of firms: 2,467.
- Panel B — Differential Effects (interaction terms):
  - NK-to-All Index yields interaction signs opposite to baseline for asset size, fixed asset ratio, and economic cooperation, suggesting inability to distinguish upside vs downside risks.
  - GPR Korea Index:
    - Log(Index) * foreigner ownership = 0.0035*** (0.0010) (similar to benchmark)
    - Other interactions not significant, suggesting the GPR Korea index focuses more on broader downside geopolitical risks and may not capture upside inter-Korea swings.
- Interpretation:
  - A simple frequency-based NK-to-All Index may confound upside and downside news.
  - Benchmark GPRNK (constructed using both negative and positive news) provides advantage as a net negative index capturing direction of geopolitical impact.

### 4.2. Robustness
- Robustness checks using restricted samples (Table 8). All columns include firm and time fixed effects; controls include price-to-book value ratio, leverage ratio, ROA; first-order terms of firm-level sensitivity measures included.
- Column (1): Exclude delisted firms as of Dec. 2019
  - Log(GPRNK) * foreigner ownership = 0.0052*** (0.0009)
  - Number of observations: 331,168; Number of firms: 1,836; Adjusted R2: 0.2074
- Column (2): Manufacturing firms only
  - Log(GPRNK) * foreigner ownership = 0.0044*** (0.0012)
  - Number of observations: 256,464; Number of firms: 1,598; Adjusted R2: 0.1752
- Column (3): Exclude January and December observations
  - Log(GPRNK) * foreigner ownership = 0.0057*** (0.0011)
  - Number of observations: 326,679; Number of firms: 2,426; Adjusted R2: 0.1603
- Column (4): Exclude extreme dependent variable values (dependent variable range restricted to -100 log-point to 100 log-point)
  - Log(GPRNK) * foreigner ownership = 0.0054*** (0.0009)
  - Number of observations: 390,336; Number of firms: 2,426; Adjusted R2: 0.2056
- Column (5): Market Return > 0
  - Log(GPRNK) * foreigner ownership = 0.0063*** (0.0014)
  - Number of observations: 216,128; Number of firms: 2,426; Adjusted R2: 0.1036
- Column (6): Market Return < 0
  - Log(GPRNK) * foreigner ownership = 0.0062*** (0.0016)
  - Number of observations: 175,277; Number of firms: 2,426; Adjusted R2: 0.1574
- Column (7): Sample restricted to Year >= 2003
  - Log(GPRNK) * foreigner ownership = 0.0052*** (0.0011)
  - Number of observations: 325,513; Number of firms: 2,379; Adjusted R2: 0.1357
  - Note: Coefficient on share of fixed assets becomes smaller and statistically insignificant in this restricted sample.
- Column (8): Sample restricted to Year >= 2012
  - Log(GPRNK) * foreigner ownership = 0.0046*** (0.0017)
  - Number of observations: 165,060; Number of firms: 2,012; Adjusted R2: 0.1248
  - Interaction coefficients significant only for foreign ownership and economic cooperation dummy in this recent-sample restriction.
- Overall robustness conclusion:
  - Main results—positive moderating role of foreign ownership and negative sensitivity for firms with larger fixed asset shares and those involved in economic cooperation with North Korea—are generally robust across restricted samples and periods, with some variation in statistical significance depending on sample restrictions.

*Source: wpiea2021251-print-pdf — 4.   Results*

### 5.   Conclusion

### 5.   Conclusion

### Development of a country-specific geopolitical risk measure
- Introduced a new measure of geopolitical risk in the context of inter-Korean relations (GPRNK).
- Index focuses more on country-specific risk factors than existing measures and captures fluctuations by accounting for both tension-increasing and tension-decreasing moments.
- Comprehensively includes North Korea-related issues such as: the threat of war, the development of WMDs, sanctions, talks, and economic cooperation.

### Dynamics of GPRNK
- The GPRNK index indicates that geopolitical tension on the Korean Peninsula was at its peak during the North Korea nuclear missile crisis of 2017.
- The index sharply increases with the occurrence of nuclear tests, missile launches, and military aggression from the North.
- The agreement to hold the first inter-Korean summit eased tensions most dramatically.
- The index also drops around the time of bilateral or multilateral talks.

### Firm-level effects on South Korean stock returns
- Heightened geopolitical risk stemming from inter-Korea relations depresses stock returns in South Korea.
- The negative effects are particularly pronounced for:
  - firms with a larger share of domestic investors,
  - firms with a higher share of fixed assets compared to their total assets,
  - firms that are involved in inter-Korea economic cooperation.
- Results point to two important mechanisms explaining firm stock-price reactions:
  - irreversibility of investment,
  - international portfolio diversification.

### Robustness and sensitivity checks
- Media selection: media sample selection is based on popularity; Table A-1 shows paid subscription rankings for newspapers in 2012 and top outlets were used.
- Economic cooperation classification: restricting the economic cooperation dummy to firms that once participated in business with North Korea (shaded cells in Table A-2) yields results very similar to the baseline, though the coefficient of GPRNK interaction with economic cooperation is slightly smaller.
- Controls for other uncertainty measures:
  - Including interactions with the VKOSPI-200 index (market volatility) still yields strongly significant coefficients for interactions with foreign ownership and the economic cooperation dummy (Table A-3). Sample covers 2003m1~2019m12.
  - Controlling for Economic Policy Uncertainty (EPU) interactions does not materially change the estimated GPRNK effects (Table A-4). Sample covers 1996m2~2019m12.
- Timing and windowing checks:
  - Three-month moving-average index calculations were used to reduce timing bias. Windows applied include [t-1, t+1] (equal weight and double weight on t) and [t-2, t] (equal weight and double weight on t). These alternative time settings yield results similar to Column (6) in Table 5.
  - The moving-averaged index implementations are explicitly: window [t-1, t+1], window [t-2, t], subscription-weighted index, and standardized index (Table A-5).

### Additional analyses and alternative constructions
- GPRNK is examined by type of news media (Figure A-2) and as a weighted index (Figure A-3).
- Comparisons made with broader economic uncertainty indices (Figure A-4).
- Alternative GPRNK constructions:
  - A “broad GPRNK” index is constructed by skipping the exclusion step for words that negate the topic.
  - An alternative GPRNK including joint exercises is computed by adding the keywords “exercises” and “condemn” (Figure A-5).

### Contribution
- The index provides a country-specific measure of geopolitical risk for the Korean Peninsula and extends existing literature on geopolitical risk measurement.
- The analysis deepens understanding of how South Korean markets react to geopolitical swings by estimating differential effects by firm-specific characteristics.

*Source: wpiea2021251-print-pdf - 5.   Conclusion (IMF working paper).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021251-print-pdf.pdf_