수출신용기관의 정책 커뮤니케이션은 주식시장에 영향을 미치는가? 한국수출입은행의 사례

Do Export Credit Agencies Move Markets? Evidence from Korea Eximbank Communication

차범석¹ · 이재호² · 손한¹^*

Bumsuk Cha¹, Jaeho Lee², Han Sun¹^*

¹ 한양대학교 컴퓨테이셔널파이낸스공학과

² 한양대학교 경영학과 재무금융전공

¹ Department of Computer Science and Financial Engineering, Hanyang University

² Department of Finance, Hanyang University Business School

^*교신저자: 손한 (mumuaitutu@hanyang.ac.kr)

^*Corresponding author: Han Sun (mumuaitutu@hanyang.ac.kr)

발행: 2026년 8월·Vol. 30, No. 3·pp. 165-180

DOI: https://doi.org/10.17287/kbr.2026.30.3.165

Abstract

초록

본 연구는 정부와 연계된 수출신용기관의 정책 커뮤니케이션이 주가에 반영되는지를 분석한다. 한국수출입은행을 대상으로 2016년부터 2024년까지 발표된 수출금융, 보증, 프로젝트 파이낸스 및 산업별 지원 관련 공식 발표에 대한 주식시장의 반응을 사건연구 방법론을 이용하여 분석하였다. 분석 결과, 주가는 발표 당일 긍정적으로 반응했으며, 일반적인 협약보다는 프로그램 출범, 유동성 지원 또는 보증 확대와 같은 구체적인 정책 조치를 포함한 발표에서 더 강한 효과가 나타났다. 이러한 효과는 수출지향적 기업과 재무적 제약이 큰 기업, 그리고 경제적 불확실성이 높은 시기에 더욱 뚜렷하게 나타났다. 이는 정책 커뮤니케이션이 시장이 중요하게 평가하는 미래 금융 여건에 관한 정보를 전달한다는 것을 보여준다.

Abstract

This study examines whether policy communication by a government-linked export credit institution is reflected in stock prices. Focusing on the Export–Import Bank of Korea, we analyze stock market reactions to official announcements on export finance, guarantees, project finance, and sectoral support from 2016 to 2024 using an event-study approach. Equity prices respond positively on announcement days, with stronger effects for announcements involving concrete policy actions such as program launches, liquidity support, or expanded guarantees rather than general agreements. The effects are more pronounced for export-oriented firms, financially constrained firms, and during periods of elevated economic uncertainty, indicating that policy communication conveys information about future financing conditions valued by the market.

주제어:수출신용기관정책 커뮤니케이션주식시장 반응무역금융

Keywords:Export credit agenciesPolicy communicationStock market reactionsTrade finance

Ⅰ. Introduction

Export Credit Agencies (ECAs) provide credit, guarantees, and insurance when private trade finance is limited or costly. Their role is particularly relevant in export-dependent economies and in sectors that rely on large overseas projects. Korea provides a useful setting because policy banks are active in strategic industries and export finance represents an important component of industrial policy.

Prior research shows that policy communication can affect asset prices through changes in beliefs and risk assessments (Hansen, McMahon, & Prat, 2018; Gómez-Cram & Grotteria, 2022; Gorodnichenko, Pham, & Talavera, 2023). Korean evidence also indicates that institutional and regulatory information affects how investors price firm-level information (Park, Lee, & Lee, 2024). Existing evidence, however, is concentrated on central banks. ECA communication differs because it concerns targeted credit allocation to identifiable firms, projects, and sectors rather than economy-wide interest rates.

The setting is the Export-Import Bank of Korea (KEXIM). KEXIM is tightly linked to the sovereign. Its legal framework includes explicit loss-compensation provisions by the government (Export-Import Bank of Korea Act, 2008). This institutional design can make KEXIM announcements informative about future support capacity and industrial priorities. KEXIM also releases frequent and detailed policy communications, which are suitable for event-based tests.

This paper extends the policy- communication literature to an institution whose announcements concern targeted credit allocation rather than economy-wide interest rates. It provides event-study evidence on whether KEXIM announcements are priced in equity markets, measures variation in announcement content and tone, and examines whether responses differ with firms’ exposure to export activity and financing constraints. These tests distinguish the setting from central-bank communication, for which the transmission channel is broader and less firm-specific.

Ⅱ. Institutional Background and Hypotheses

ECAs address frictions in cross-border contracting and trade finance. Private lenders often face high screening costs and limited information about foreign buyers and large overseas projects. These frictions shape payment terms and credit provision in international trade (Antras & Foley, 2015). ECAs mitigate such frictions with direct lending, guarantees, and political-risk insurance. Their role becomes more salient when risk appetite collapses and private credit tightens. Public credit programs and guarantees can relax constraints and expand credit supply when private provision is weak (Bachas, Kim, & Yannelis, 2021).

KEXIM is a salient ECA because its mandate and sovereign linkage are explicit. Article 37 of the Export-Import Bank of Korea Act states that the bank covers net losses with reserves, and the government compensates losses when reserves are insufficient (Export-Import Bank of Korea Act, 2008). KEXIM’s operational scale is also large. In 2022, it reported total funding of KRW 76.7 trillion, exceeding its target by 11% (Korea Eximbank, 2022). This combination implies that KEXIM announcements can plausibly shift expectations about sector-level financing conditions and project viability.

The key mechanism is informational. A large literature shows that policy communication can move asset prices by changing beliefs, reducing uncertainty, and revealing private assessments. Text and deliberation content are informative about policy stances (Hansen, McMahon, & Prat, 2018). Verbal tone can also contain incremental signals that markets price (Gómez-Cram & Grotteria, 2022; Gorodnichenko, Pham, & Talavera, 2023). Communication can matter even when it does not immediately change realized policy actions because it changes expectations and perceived tail risk. Evidence from Korea further shows that financing-related corporate events are reflected in shareholder valuations (Park, 2022), while policy uncertainty is associated with both asset prices and firm outcomes (Baker, Bloom, & Davis, 2016).

However, the evidence base is heavily concentrated on central banks. ECAs differ in two ways. First, their communications are not about economy-wide discount rates. They are about targeted credit allocation and industrial policy priorities. Second, ECAs can implicitly certify project quality through due diligence and screening. This can reduce information asymmetry for investors. Despite these channels, rigorous asset-pricing evidence for ECA communication is scarce. Existing work on export-credit institutions often focuses on real outcomes or institutional disruptions, rather than the information content of announcements (Benmelech & Monteiro, 2023).

These differences motivate firm-level hypotheses.

H1. KEXIM announcements of new or expanded support are associated with positive abnormal returns for exposed firms.
H2. Abnormal returns are larger when announcements are more supportive in tone or contain more concrete resource commitments.
H3. Effects are stronger for firms that are more export-oriented, more financially constrained, and less able to rely on internal capital markets.
H4. Effects are stronger when macro or policy uncertainty is elevated.

Ⅲ. Data and Empirical Framework

3.1 Sample and Data Sources

Our sample covers January 1, 2016 to December 31, 2024. We start from all firms listed on the Korea Exchange (KOSPI and KOSDAQ). We exclude financial firms and regulated utilities. We keep firms with daily price data and annual accounting data from FnGuide. The resulting universe contains 1,636 firms.

We define the two key cross-sectional variables explicitly. Export is the firm’s export-to-sales ratio at fiscal year-end t-1, computed from FnGuide segment-disclosure data, and is used as a continuous variable in the main regressions. For descriptive splits and figures we additionally use a binary indicator HighExport, which equals one for firms whose Export ratio is above the cross-sectional median in year t-1 (median ≈ 0.379; mean ≈ 0.396). FC is a binary financial-constraint indicator constructed following Hadlock and Pierce (2010): we compute the firm-year HP/SA index as -0.737×Size + 0.043×Size2 - 0.040×Age (with Size = log (total assets) winsorized at log(4.5 billion KRW) and Age in years since listing, capped at 37), and assign FC = 1 to firms in the top tercile of the index in year t-1 and FC = 0 otherwise. By construction, FC is more common among smaller and younger firms, which accounts for the strong negative correlation between FC and Size in Table 3 ($\rho \approx -0.79$).

<Table 3> Correlations

	CAR[-1,1]	TONE	Export	FC	Size	Lev	BM
CAR[-1,1]	1.000	0.034	0.003	0.001	-0.003	-0.004	0.001
TONE	0.034	1.000	0.001	-0.000	-0.000	-0.001	-0.001
Export	0.003	0.001	1.000	0.010	-0.011	-0.032	-0.014
FC	0.001	-0.000	0.010	1.000	-0.786	0.013	0.011
Size	-0.003	-0.000	-0.011	-0.786	1.000	-0.028	-0.004
Lev	-0.004	-0.001	-0.032	0.013	-0.028	1.000	0.017
BM	0.001	-0.001	-0.014	0.011	-0.004	0.017	1.000

Notes. This table reports Pearson correlations among the key variables used in the empirical analysis. Variables are defined as in Table 2. Return correlations are typically small at daily frequency in short-window event studies.

We assemble three datasets. First, we obtain daily stock prices and firm fundamentals from FnGuide, which allows us to compute returns and lagged controls consistently across KOSPI and KOSDAQ firms. Second, we hand-collect KEXIM policy announcements from the official KEXIM website and verify major releases using postings by the Ministry of Economy and Finance (MOEF), the Financial Services Commission (FSC), and mainstream financial newswires. For every candidate release, we search MOEF and FSC press-room postings and at least two major financial newswires within a [-1,+1] calendar-day window. Releases posted after the KOSPI close or on non-trading days are shifted to the next trading day, with external timestamps used as a tie-breaker. Administrative notices and candidates that cannot be corroborated by at least one external source are dropped. This procedure removes 38 candidate items and yields 521 event dates. Third, we quantify announcement tone using KOSELF and a finance-domain dictionary approach following standard financial text-analysis practice (Loughran & McDonald, 2011). The daily panel is used to estimate abnormal returns, while the main regressions use a stacked firm-announcement sample. Table 1 summarizes the manually coded, mutually exclusive announcement categories.

<Table 1> Classification of KEXIM Policy Announcements

Type	Label	Coding rule (what qualifies)	Typical content keywords	Expected most-affected firms
PROGRAM	Program launch / expansion	New facility, increased ceiling, broadened eligibility, new policy package	“launch”, “expand”, “increase limit”, “new facility”, “support package”	Exporters; firms in targeted sectors; constrained firms
GUAR	Guarantee / insurance support	Guarantees, trade credit insurance, risk coverage, loss-sharing schemes	“guarantee”, “insurance”, “risk coverage”, “loss-sharing”, “policy guarantee”	SMEs; high-risk exporters; firms with weaker collateral
PF	Project finance / overseas investment	Named overseas project financing, buyer’s credit, EPC/PPP financing support	“project finance”, “buyer’s credit”, “EPC”, “overseas project”, “PPP”	Firms with overseas EPC/investment exposure
SECTOR	Sector-targeted support	Explicit sector or strategic-industry package (shipbuilding, batteries, defense, semiconductors, etc.)	“strategic industry”, sector name, “supply chain”, “high-tech”	Firms in the named sector(s)
LIQ	Working capital / liquidity	Short-term liquidity, emergency funding, rollover, maturity extension	“working capital”, “liquidity”, “emergency”, “rollover”, “maturity extension”	Firms facing near-term funding stress
AGREE	MoU / partnership / framework agreement	Agreements without immediate financing size (MoU, cooperation, framework)	“MoU”, “cooperation”, “framework agreement”	Usually weaker, more diffuse effects
GOV	Government-backed capacity change	Capital injection, policy fund allocation, statutory/mandate updates	“capital injection”, “policy fund”, “government support”, “mandate”	Broad; stronger when capacity constraints bind

Notes. This table defines the announcement-type taxonomy used to construct the event-level content variables. Each KEXIM announcement is assigned a primary category based on its dominant policy content: PROGRAM (program launch/expansion), GUAR (guarantee/insurance support), PF (project finance/overseas investment finance), SECTOR (sector-targeted support), LIQ (liquidity/working-capital support), AGREE (MoU/partnership/framework agreements without a clearly quantified facility), and GOV (government-backed capacity or mandate changes, such as capital injections or policy-fund allocations). If multiple KEXIM releases occur on the same calendar date, we merge them into a single event-day document and assign the primary category according to the dominant component of the merged text. Categories are mutually exclusive at the event-day level.

3.2 Abnormal Returns and Event Windows

We measure price reactions using abnormal returns. Let (e) denote the KEXIM announcement date, aligned to the relevant trading day. For each firm (i), we estimate a market model over a pre-event estimation window:

$$R_{i,t} = \alpha_i + \beta_i R_{m,t} + u_{i,t}$$(1)

Abnormal return is the fitted residual:

$$\mathit{AR}_{i,t} = R_{i,t} - \left(\widehat{\alpha}_{\imath} + \widehat{\beta}_{\imath} R_{m,t}\right)$$(2)

and cumulative abnormal return over an event window [a,b] is:

$$\mathit{CAR}_{i,e}[a,b] = \sum\nolimits_{\tau=a}^{b} \mathit{AR}_{i,e+\tau}$$(3)

We focus on three horizons. The main announcement window is $\mathit{CAR}_{i,e}[-1,1]$. This window allows limited leakage and short-run adjustment. To address timing uncertainty from after-close releases, we also use $\mathit{CAR}_{i,e}[0,1]$. As a placebo pre-trend check, we compute $\mathit{CAR}_{i,e}[-3,-1]$. In the absence of information leakage, this pre-event window should be close to zero on average.

3.3 Empirical Specification

The main tests use a stacked firm–announcement design. For each announcement date e, we construct an event-window outcome Yi,ew for each firm i, where

$$Y_{i,e}^{w} \in \{\, \mathit{CAR}_{i,e}[-1,1],\ \mathit{CAR}_{i,e}[0,1],\ \mathit{CAR}_{i,e}[-3,-1] \,\}$$(4)

We estimate:

$$\begin{aligned} & Y_{i,e}^{w} = \alpha + \beta_1\, \mathit{TONE}e + \beta_2'\, \mathit{TYPE}e + \gamma'\, X_{i,e-1} \\ & + \eta_i + \theta_{ym(e)} + \varepsilon_{i,e} \end{aligned}$$(5)

$\mathit{TONE}_e$ measures announcement tone, $\mathit{TYPE}_e$ captures announcement categories, and $X_{i,e-1}$ includes lagged firm controls. We include firm fixed effects $\eta_i$ and year-month fixed effects $\theta_{ym(e)}$. Because announcements are common shocks across firms, we cluster standard errors by firm and event date (Petersen, 2009).

3.4 Descriptive Statistics

Table 2 reports descriptive statistics. The sample contains 1,636 KRX firms and 521 KEXIM event dates. PROGRAM and SECTOR are the most frequent announcement categories. TONE has a mean close to zero and a standard deviation of 0.21, providing substantial cross-event variation. The mean Export ratio is 0.40, with an interquartile range of 0.25 to 0.54, and 40% of firm-years are classified as financially constrained under the SA-index tercile rule. Short-window abnormal returns are centered near zero, and the pre-event placebo window also has a mean close to zero.

<Table 2> Descriptive Statistics

Panel A. Firm characteristics (N = 1,636)
	Count	Mean	Std	25%	50%	75%
Size	1,636	13.476	1.188	12.690	13.485	14.291
Lev	1,636	0.444	0.177	0.327	0.447	0.566
BM	1,636	0.917	0.553	0.515	0.891	1.297
Export	1,636	0.396	0.196	0.247	0.379	0.537
FC	1,636	0.400	0.490	0.000	0.000	1.000
Panel B. Announcement types (N = 521)
		N		Percent
PROGRAM		106		20.3
GUAR		78		15.0
PF		47		9.0
SECTOR		103		19.8
LIQ		49		9.4
AGREE		81		15.5
GOV		57		10.9
Panel C. Abnormal-return measures (firm–announcement obs. N = 852,356)
	Count	Mean	Std	25%	50%	75%
AR[0]	852,356	0.000	0.030	-0.020	0.000	0.021
CAR[0,1]	852,356	0.001	0.040	-0.027	0.001	0.028
CAR[-1,1]	852,356	0.001	0.050	-0.033	0.001	0.035
CAR[-3,-1]	852,356	-0.000	0.050	-0.034	-0.000	0.034
Panel D. Announcement tone (N = 521)
	Count	Mean	Std	25%	50%	75%
TONE	521	0.000	0.210	-0.142	0.000	0.142

Notes. This table reports summary statistics for the firm-announcement sample used in the analysis. Panel A reports firm-level characteristics; Panel B reports announcement-type frequencies; Panel C reports abnormal-return measures in decimal units; and Panel D reports the announcement-level tone measure. Abnormal returns follow standard market-model conventions.

3.5 Correlation Analysis

Table 3 reports pairwise correlations among the key variables. Correlations between short-window abnormal returns and text tone are small in magnitude. FC is strongly negatively correlated with Size, reflecting the construction of the financial-constraint indicator.

3.6 Hypothesis Tests

3.6.1 Tests of H1 and H2

Table 4 examines H1 and H2. Relative to AGREE, PROGRAM, LIQ, SECTOR, and PF announcements have larger short-window abnormal returns. The coefficient on TONE is positive and statistically significant in all three specifications.

<Table 4> Announcement Effects: Tone and Content (H1–H2)

	(1) AR[0]	(2) CAR[0,1]	(3) CAR[-1,1]
TONE	0.110*** (27.68)	0.140*** (29.21)	0.190*** (28.89)
PROGRAM	0.060*** (4.93)	0.100*** (7.07)	0.090*** (4.63)
GUAR	0.050*** (3.73)	0.040*** (2.77)	0.050** (2.33)
PF	0.040** (2.49)	0.070*** (4.14)	0.110*** (5.01)
SECTOR	0.070*** (5.77)	0.080*** (5.32)	0.100*** (5.58)
LIQ	0.100*** (7.17)	0.130*** (7.20)	0.110*** (4.82)
GOV	0.070*** (4.46)	0.030* (1.66)	0.050** (2.26)
Firm FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Controls	Yes	Yes	Yes
N	852,356	852,356	852,356
Within R²	0.0012	0.0013	0.0012

Notes. Dependent variables are multiplied by 100 to express percentage returns. The table reports coefficient estimates, with t-statistics in parentheses. TONE is the announcement-level tone measure. Announcement-type coefficients are reported relative to AGREE. Controls include Size, Lev, BM, Export, and FC, all measured at year t-1. Firm and year fixed effects are included as indicated. Standard errors are clustered by firm and event date. ***, **, and * denote significance at the 1%, 5%, and 10% levels.

Table 4 also shows that the largest coefficients on CAR[-1,1] are associated with LIQ and PF (11 bps), followed by SECTOR (10 bps) and PROGRAM (9 bps), each measured relative to AGREE. The ranking is consistent with more concrete announcements carrying more pricing-relevant information, but the specification does not separately identify credit-supply, certification, or project-completion channels. Accordingly, the cross-type differences are interpreted as descriptive evidence on announcement content rather than as direct tests of specific mechanisms. Prior research similarly shows that policy communication containing information about future actions can affect market prices (Hansen, McMahon, & Prat, 2018; Gómez-Cram & Grotteria, 2022).

3.6.2 Validity Check: Placebo Pre-window

Table 5 reports economically small and statistically insignificant coefficients in the pre-event window, providing little evidence of systematic pre-event leakage.

<Table 5> Placebo Test Using CAR[-3,-1]

	CAR[-3,-1]
TONE	0.000 (0.59)
PROGRAM	0.000 (-0.04)
GUAR	0.000 (-0.16)
PF	0.000 (-0.19)
SECTOR	-0.020 (-1.01)
LIQ	0.010 (0.32)
GOV	-0.010 (-0.53)
Firm FE	Yes
Year FE	Yes
Controls	Yes
N	852,356
Within R²	0.0000

Notes. The dependent variable is CAR[-3,-1] multiplied by 100 to express percentage returns. The table reports coefficient estimates, with t-statistics in parentheses. Announcement-type coefficients are reported relative to AGREE. Controls include Size, Lev, BM, Export, and FC, all measured at year t-1. Firm and year fixed effects are included as indicated. Standard errors are clustered by firm and event date. ***, **, and * denote significance at the 1%, 5%, and 10% levels.

3.6.3 Test of H3: Cross-sectional Heterogeneity

Table 6 reports positive and statistically significant coefficients on TONE×Export and TONE×FC, indicating that the association between announcement tone and CAR[-1,1] is stronger among firms with greater export exposure and among financially constrained firms.

<Table 6> Heterogeneity Tests (H3), Dependent Variable: CAR[-1,1]

	(1) Baseline	(2) Export	(3) FC	(4) Both
TONE	0.190*** (28.89)	0.130*** (10.57)	0.150*** (20.99)	0.100*** (7.29)
TONE×Export		0.140*** (5.16)		0.140*** (5.15)
TONE×FC			0.100*** (9.21)	0.100*** (9.26)
Firm FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
Controls	Yes	Yes	Yes	Yes
N	852,356	852,356	852,356	852,356
Within R²	0.0012	0.0013	0.0013	0.0014

Notes. The dependent variable is CAR[-1,1] multiplied by 100 to express percentage returns. The table reports coefficient estimates, with t-statistics in parentheses. Export and FC are measured at year t-1. Controls include Size, Lev, BM, Export, and FC, all measured at year t-1. Firm and year fixed effects are included as indicated. Standard errors are clustered by firm and event date. ***, **, and * denote significance at the 1%, 5%, and 10% levels.

3.6.4 Test of H4: Stress/Uncertainty Periods

Table 7 reports a positive and statistically significant TONE×COVID coefficient, indicating a stronger association between announcement tone and CAR[-1,1] during the COVID stress period.

<Table 7> Stress-Period Amplification (H4), Dependent Variable: CAR[-1,1]

	(1) Baseline	(2) COVID
TONE	0.190*** (28.89)	0.170*** (26.35)
TONE×COVID		0.120*** (8.07)
Firm FE	Yes	Yes
Year FE	Yes	Yes
Controls	Yes	Yes
N	852,356	852,356
Within R²	0.0012	0.0013

Notes. The dependent variable is CAR[-1,1] multiplied by 100 to express percentage returns. The table reports coefficient estimates, with t-statistics in parentheses. COVID identifies the COVID-period stress window. Controls include Size, Lev, BM, Export, and FC, all measured at year t-1. Firm and year fixed effects are included as indicated. Standard errors are clustered by firm and event date. ***, **, and * denote significance at the 1%, 5%, and 10% levels.

3.6.5 Robustness to Bank of Korea Communication Overlap

Potential overlap with Bank of Korea (BOK) Monetary Policy Committee communication is examined in Table 8. BOKDAY equals one when a KEXIM event falls within ±1 calendar day of a scheduled MPC rate decision or policy statement; 67 of 521 event dates meet this condition. Excluding these dates changes the TONE coefficient only from 0.190 to 0.200. In the full sample, the TONE×BOKDAY interaction is small and statistically insignificant. The baseline estimate is therefore not materially altered by excluding nearby BOK communication dates.

<Table 8> Robustness to BOK Communication Overlap (Dependent Variable: CAR[-1,1])

	(1) Full sample	(2) Clean (no BOK overlap)	(3) Full + BOK controls
TONE	0.190*** (28.89)	0.200*** (26.50)	0.190*** (28.21)
BOKDAY			-0.020 (-0.85)
TONE×BOKDAY			-0.010 (-0.32)
Type controls	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Controls	Yes	Yes	Yes
Events	521	454	521
N	852,356	742,917	852,356
Within R²	0.0012	0.0012	0.0012

Notes. The dependent variable is CAR[-1,1] multiplied by 100 to express percentage returns. The table reports coefficient estimates, with t-statistics in parentheses. BOKDAY equals one for KEXIM event dates within ±1 calendar day of scheduled BOK MPC communication. Controls include Size, Lev, BM, Export, and FC, all measured at year t-1. Firm and year fixed effects are included as indicated. Standard errors are clustered by firm and event date. ***, **, and * denote significance at the 1%, 5%, and 10% levels.

3.6.6 Alternative Event Windows

Table 9 re-estimates the baseline specification using CAR[-2,+2], CAR[0,+2], and CAR[-1,+2]. The TONE coefficient remains positive and significant at the 1% level, ranging from 0.170 to 0.210 in percentage-return units. PROGRAM, LIQ, SECTOR, and PF also retain positive coefficients across the alternative windows. The reported estimates are therefore similar across the short event windows.

<Table 9> Robustness to Alternative Event Windows

	CAR[-1,+1]	CAR[-2,+2]	CAR[0,+2]	CAR[-1,+2]
TONE	0.190*** (28.89)	0.210*** (27.45)	0.170*** (28.10)	0.200*** (28.50)
PROGRAM	0.090*** (4.63)	0.110*** (5.04)	0.090*** (5.11)	0.100*** (4.85)
LIQ	0.110*** (4.82)	0.130*** (5.12)	0.120*** (5.34)	0.120*** (5.05)
SECTOR	0.100*** (5.58)	0.110*** (5.79)	0.090*** (5.42)	0.100*** (5.61)
PF	0.110*** (5.01)	0.120*** (5.27)	0.080*** (4.30)	0.110*** (5.06)
Other types	Yes	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
Controls	Yes	Yes	Yes	Yes
N	852,356	852,356	852,356	852,356

Notes. Dependent variables are multiplied by 100 to express percentage returns. The table reports coefficient estimates, with t-statistics in parentheses. Announcement-type coefficients are reported relative to AGREE. Controls include Size, Lev, BM, Export, and FC, all measured at year t-1. Firm and year fixed effects are included as indicated. Standard errors are clustered by firm and event date. ***, **, and * denote significance at the 1%, 5%, and 10% levels.

3.6.7 Market Stress Indices: VIX and Bank of Korea Financial Stress Index

We extend the stress-period analysis beyond the COVID indicator using two market-based measures: the Cboe Volatility Index (VIX), measured on the trading day preceding each KEXIM event, and the Bank of Korea Financial Stress Index (FSI), using the most recent monthly observation available before the event. For ease of interpretation, HighVIX identifies events in the top tercile of the 2016–2024VIX distribution (threshold ≈ 22.4), while HighFSI identifies observations in the BOK caution/crisis range (threshold ≈ 12). Table 10 reports positive and statistically significant interactions of TONE with both indicators. The estimates are 0.140 for TONE×HighVIX and 0.110 for TONE×HighFSI in the separate specifications; both remain positive when included jointly. These results show that the tone-return association is stronger in higher-stress periods under both market-based measures.

<Table 10> Stress-Period Amplification with Market-Based Stress Indices (Dependent Variable: CAR[-1,1])

	(1) HighVIX	(2) HighFSI	(3) Both
TONE	0.170*** (25.10)	0.180*** (25.92)	0.160*** (23.45)
TONE×HighVIX	0.140*** (7.84)		0.110*** (6.05)
TONE×HighFSI		0.110*** (6.42)	0.080*** (4.21)
Type controls	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Controls	Yes	Yes	Yes
N	852,356	852,356	852,356
Within R²	0.0013	0.0013	0.0013

Notes. The dependent variable is CAR[-1,1] multiplied by 100 to express percentage returns. The table reports coefficient estimates, with t-statistics in parentheses. HighVIX identifies events in the top tercile of the VIX distribution, and HighFSI identifies events in the BOK FSI caution/crisis range. Controls include Size, Lev, BM, Export, and FC, all measured at year t-1. Firm and year fixed effects are included as indicated. Standard errors are clustered by firm and event date. ***, **, and * denote significance at the 1%, 5%, and 10% levels.

Ⅳ. Conclusion

Announcement-day returns are positively associated with the tone and content of KEXIM communication, with larger estimates for export-intensive and financially constrained firms and during higher-stress periods. The pre-event placebo and the reported robustness tests yield similar inferences across alternative event windows and after excluding dates near BOK com- munication. The findings are consistent with KEXIM announcements containing valuation-relevant information, but the daily-frequency design does not isolate individual transmission mechanisms or fully remove concurrent policy and macroeconomic news. Intraday data and direct links between announcements, financing flows, and real activity would allow these channels to be examined more directly.

<Teaching Note> Do Export Credit Agencies Move Markets? Evidence from Korea Eximbank Communication

1. Synopsis

This teaching case examines stock-market responses to policy communication by an Export Credit Agency (ECA). Using announcements by the Export-Import Bank of Korea (KEXIM), it focuses on how information about prospective credit, guarantees, project finance, and sectoral support is incorporated into equity prices.

Because ECAs allocate policy-backed finance to identifiable firms and industries, their announcements may alter expectations about financing conditions before funds are disbursed. The case uses this setting to discuss event-study design, heterogeneous exposure, and the interpretation of policy-linked market reactions.

2. Learning Objectives

The teaching case has five learning objectives:

Explain how export credit agencies differ from central banks in the scope and allocation of policy support.

Assess why an announcement may affect asset prices before any lending or guarantee is executed.

Interpret short-window abnormal returns and distinguish statistical evidence from mechanism claims.

Compare market responses across firms with different export exposure and financing constraints.

Evaluate whether observed price reactions are consistent with information effects, policy support, or both.

3. Target Audience & Courses

Target Audience

Upper-level undergraduate students in finance, economics, or public policy.
MBA and Master’s students studying financial markets, trade finance, or industrial policy.

Applicable Courses

Corporate Finance
Financial Markets and Institutions
International Finance and Trade Finance
Government and Industrial Policy
Case-based Finance or Economics courses using event-study applications

4. Case Questions

1. Through which information channels could a KEXIM announcement affect equity prices before financing is disbursed?
2. Which announcement characteristics are associated with larger short-window returns in the case evidence?
3. How should the stronger estimates for export-oriented and financially constrained firms be interpreted?
4. What does the stress-period evidence add to the baseline results, and what does it not establish?
5. Which additional data or research designs would be needed to distinguish information, certification, and credit-supply mechanisms?

5. Case Analyses & Key Concepts

Analytical Approach

Short-window event-study design
Firm-announcement regressions with fixed effects
Cross-sectional heterogeneity, placebo, and robustness analyses

Key Concepts

Policy communication and expectation formation
Information asymmetry and government-backed credit
Targeted policy transmission
Limits of mechanism inference from short-window event studies

The case is designed to separate the empirical finding that KEXIM announcements are associated with short-window equity-price responses from stronger claims about the channels generating those responses.