## _wp1482

## Source details

**Canonical URL:** [_wp1482](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1482.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1482.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1482.pdf.json)

---

### Introduction and motivation
- A record one billion tourists crossed international borders in 2012 (UNWTO).
- WTTC estimates travel and tourism accounts for 9 percent of global GDP from “direct and indirect activities combined”.
- In 2006–10, international tourism receipts represented about 6 percent of international trade of goods and services, and nearly 2 percent of the world’s GDP.
- International trade in fuels accounts for 10 percent of total trade; international remittances stand at ¾ percent of the world’s GDP.
- In 45 of the 191 countries and territories with 2009 data, international tourism accounted for over 20 percent of total exports.
- Several OECD countries (Australia, Greece, New Zeeland, Portugal, Spain, Turkey) derive 14 to 25 percent of export earnings from foreign tourists; US, France and Italy are in the 8 to 10 percent range.
- The paper analyzes determinants of international tourism flows and cross-border spillovers using an extensive panel from UNWTO raw data covering the universe of bilateral tourism flows over a decade.

### Data and dataset construction
- Primary source: UNWTO raw data, supplemented by WDI, IFS, Penn World Tables (PWT), DOTS, and CEPII.
- UNWTO electronic data distributed in five-year blocks; two blocks available for 1999–2004 and 2005–2009; as of early 2014 full UNWTO dataset covers 1995 through 2012.
- Key measurement heterogeneities:
  - Countries differ on counting at border vs. hotel, tourists vs. visitors, origin determined by nationality vs. residence.
  - Tourist-nights collected by about 40 percent of all countries.
  - Many statistical offices aggregate origin countries (e.g., “Other Africa”, “Benelux”).
- Real exchange rate measures used:
  - (i) Bilateral real exchange rate (RER) from IFS.
  - (ii) Real effective exchange rates (REER) of origin and destination from IFS.
  - (iii) PWT’s PPP factors for origin and destination (bilateral PPP vis-à-vis the United States).
  - (iv) PPP factor ratio (destination PPP factor divided by origin PPP factor), interpreted as an appreciation in dollar terms when it increases.

### Dataset summary statistics (selected)
- Unbalanced panel contains 128,304 observations (each = number of tourists from origin to destination in a given year).
- Tourist arrivals dataset:
  - UNWTO full dataset: 128,304 observations; 17,441 country-pairs; 210 destinations; Total tourist arrivals, millions: 8,475; Observations kept, %: 100.0; Arrivals kept, %: 100.0
  - Country of origin unambiguously identified: 118,115 observations; 15,839 country-pairs; 204 destinations; Total tourist arrivals, millions: 7,812; Observations kept, %: 92.1; Arrivals kept, %: 92.2
  - Key variables (GDPs, distance) available: 103,676 observations; 13,573 country-pairs; 173 destinations; Total tourist arrivals, millions: 7,678; Observations kept, %: 80.8; Arrivals kept, %: 90.6
  - Minimum 100 tourists annually: 67,673 observations; 7,966 country-pairs; 173 destinations; Total tourist arrivals, millions: 7,672; Observations kept, %: 52.7; Arrivals kept, %: 90.5
- Tourist-nights dataset:
  - UNWTO full dataset: 32,152 observations; 4,208 country-pairs; 93 destinations; Total tourist-nights, millions: 9,460; Observations kept, %: 100.0; Tourist-nights kept, %: 100.0
  - Country of origin unambiguously identified: 28,315 observations; 3,685 country-pairs; 83 destinations; Total tourist-nights, millions: 7,321; Observations kept, %: 88.1; Tourist-nights kept, %: 89.0
  - Key variables (GDPs, distance) available: 24,904 observations; 3,139 country-pairs; 71 destinations; Total tourist-nights, millions: 7,064; Observations kept, %: 77.5; Tourist-nights kept, %: 87.7
  - Minimum 100 tourists annually: 19,781 observations; 2,391 country-pairs; 71 destinations; Total tourist-nights, millions: 7,062; Observations kept, %: 61.5; Tourist-nights kept, %: 87.7

### Empirical approach and scope
- Framework: gravity model for bilateral tourism arrivals and tourist-nights; compares specifications with merchandise trade.
- Introduced method to include bilateral trade flows as independent variable in gravity estimations for non-merchandise flows.
- Examines supply-side and demand-side determinants and impacts on arrivals and average length of stay.
- Presents tourist-nights analysis (smaller dataset) and focused examination of small island destinations.
- Core estimators: Traditional panel, Country fixed effects (CFE), Country-pair fixed effects (CPFE)/demeaned OLS, First-differences (FD); Random effects (RE) and Hausman-Taylor (HT) used to recover time-invariant coefficients under assumptions.

### Key empirical findings — fit and gravity variables
- Gravity equation fit:
  - Performs very well for bilateral tourism arrivals; for some specifications fit for tourism is better than for goods trade.
  - Fit considerably better than reported for FDI and remittances flows.
- Macroeconomic determinants:
  - Elasticities of bilateral tourism with respect to origin and destination GDPs are large, though smaller than near-unit elasticities estimated for goods trade.
  - Strong trade ties positively associated with higher tourism flows, indicating a considerable share of business travelers, especially intra-OECD.
  - Real exchange rate effects: appreciation of origin currency increases bilateral tourism; appreciation of destination currency reduces it. Elasticity is large and robust across techniques and RER measures.
- Traditional gravity variables:
  - Distance negative with elasticity nearly identical to merchandise flows (examples: tourism "-1.589***" vs trade "-1.541***" in extended CFE specifications).
  - Language ties more important for tourism than merchandise trade.
  - Historical colonial relationships less important for tourism than trade.
  - Regional trade agreements associated with a small positive effect (e.g., tourism "0.252***" vs trade "0.187***" in extended CFE).
  - Common currency results mixed: reported "0.177***" in some tables but discussion notes negative coefficient in some models driven by Eurozone; non-Euro currency unions show positive coefficients between "0.3" and "0.6" depending on specification.
  - Economic remoteness of destination associated with higher tourism flows (premium on off-the-beaten-path destinations).

### Non-traditional demand- and supply-side determinants
- Supply-side:
  - Direct flight presence positively associated with tourism, but reverse causality dominates (direct flight often added the year after bilateral tourism increases).
  - Number of hotel rooms in destination positively associated with tourism inflows; reverse causality does not seem to play a role for hotel rooms.
  - Both supply variables less important for intra-OECD tourism flows.
- Non-traditional demand-side:
  - Tourists prefer regions with similar climates and prefer warmer countries.
  - Time difference (jet lag) has a negative impact on tourism flows.
  - Cultural capital (UNESCO World Heritage sites) plays a role.
  - Tourists avoid countries with ongoing conflicts.

### Estimation results — tourism versus merchandise trade (CFE comparisons)
- Selected extended-specification CFE coefficients (Table 2 examples):
  - Log Origin GDP: tourism "0.557***" (regression 2) vs trade "0.946***" (regression 6).
  - Log Destination GDP: tourism "1.040***" vs trade "0.924***".
  - Log weighted distance elasticity: tourism "-1.589***" vs trade "-1.541***".
  - Common border: tourism "0.201***" vs trade "0.427***".
  - Members of regional trade agreement: tourism "0.252***" vs trade "0.187***".
- Goodness of fit:
  - Tourism extended CFE (regression 2): Observations "104,627"; R^2 "0.848".
  - Trade extended CFE (regression 6): Observations "216,038"; R^2 "0.733".
- Qualitative interpretations:
  - Gravity model explains tourism and merchandise trade similarly; distance elasticities comparable (tourism "1.59" vs trade "1.54").
  - Origin-income elasticity for tourism is much lower than for goods trade; intra-OECD tourism elasticity rises to just above "1" but remains below goods trade elasticity for the same group.
  - Contiguity effect: common border increases tourists by "150 percent" for tourism (exp(coef)-1 interpretation) vs about "50 percent" for goods trade.
  - Common-currency puzzle: negative common-currency coefficients in some specifications are driven by Eurozone countries; non-Euro unions show positive effects.

### Non-traditional determinants — RE, CFE, HT selected results (reduced sample, min 100 tourists)
- Sample size: Observations "67,673"; Country-pairs "7,966".
- Selected coefficients (Table 3 summaries):
  - Log Origin GDP: RE "0.658***"; CFE "0.452***"; HT "0.516***".
  - Log Destination GDP: RE "0.589***"; CFE "1.177***"; HT "1.170***".
  - Log weighted distance: RE "-1.292***"; CFE "-1.614***"; HT "-1.013***".
  - Common currency: RE "0.071*"; CFE "-0.088***"; HT "0.028".
  - Members of regional trade agreement: RE "0.074***"; CFE "0.298***"; HT "0.012".
  - Log Bilateral trade (residuals): RE "0.057***"; HT "0.054***".
  - Log PPP Factor ratio (destination/origin): RE "-0.157***"; HT "-0.202***".
    - Interpretation: a "1 percent" real appreciation of the destination vis-à-vis the origin reduces arrivals by around "0.16–0.20 percent" across methods; CFE implies around "0.18 percent".
  - Direct flight dummy: RE "0.272***"; CFE "0.869***"; HT "0.201***" (implied impacts range from "20 percent" to "80 percent" depending on method).
  - Log Destination Hotel rooms: RE "0.265***"; CFE "0.143***"; HT "0.124***".
  - Time difference in hours: RE "-0.044***"; CFE "-0.018***"; HT "-0.062***".
  - Climate Similarity Index: RE "0.367***"; CFE "0.221***"; HT "0.375***".
  - Destination conflict magnitude: RE "-0.079***"; CFE "-0.083***"; HT "-0.056***".
  - Origin World Heritage sites in 2011: RE "0.014***"; CFE "0.023***".
  - Destination World Heritage sites in 2011: RE "0.012***"; HT "-0.004" (HT not significant).
- R^2 ranges reported across specifications: within "0.308"–"0.317"; between "0.517"–"0.681"; overall up to "0.829" depending on model.

### Real exchange rate and price sensitivity (detailed)
- Bilateral RER elasticity:
  - Around "-0.2" (examples: Log Bilateral RER -0.186*** to -0.213*** across specs).
- Origin vs destination RER (FD/specifications splitting effects):
  - Origin RER/PPP factor: positive and larger — examples Log Origin PPP factor "0.266***"; Log Origin REER ~"0.274***".
  - Destination RER/PPP factor: negative and smaller — examples Log Destination PPP factor "-0.084***"; Log Destination REER ~"-0.106***" to "-0.156***".
  - Summary: origin’s RER elasticity close to "0.27–0.30"; destination’s RER elasticity around "0.14–0.15" in fully controlled specifications.
- Interpretation:
  - Reverse causality biases destination coefficient downward; origin coefficient (~0.28) may be an upper bound for true destination effect.
  - Intra-OECD tourism: RER elasticities about twice higher than world sample; destination elasticity around "0.35–0.4".
- Magnitude implication: a ten percent real depreciation of the destination associated with roughly a 1.5 percent increase in tourism arrivals when using destination elasticity ≈ "0.15".

### Tourist-nights, duration, and margins
- Tourist-nights respond through both arrivals and average duration of stay.
- Tourist-nights more sensitive to destination RER than arrivals; intensive margin (duration) important.
  - Example intensive margin on destination PPP misalignment ≈ "0.41" (difference 0.744–0.333 in one specification).
- Average stay regressions:
  - Origin GDP and origin RER increase travel incidence but not length of stay.
  - Destination RER appreciation leads to shorter stays.
  - Growing destination GDP associated with shorter average stays (linked to higher share of business travel).
  - Stronger trade connections associated with shorter stays (business travel effect).
- Combined effects: some specifications imply overall impact of a 1 percent change in destination RER on tourism flows exceeds "0.7" when measured in tourist-nights; a bilateral RER-focused regression reports elasticity "0.45" (Table 8 discussion).

### Supply-side: direct flights and hotel capacity (dynamics and causality)
- Direct flight:
  - CPFE (full sample) presence associated with ~22 percent higher bilateral tourism flows (computed from CPFE coefficient ~0.2).
  - FD shows contemporaneous addition of a flight associated with a "0.040***" coefficient (4 percent increase).
  - Leads and lags significant; evidence that flights are established in response to rising tourism (reverse causality).
  - Small islands: marginal effect stronger — interaction implies marginal effect of "0.1" (10 percent) with s.e. "0.03" (significant at 1 percent).
- Hotel rooms:
  - Positive and significant associations: CPFE/FD coefficients e.g., "0.138***", "0.246***".
  - FD often shows larger short-run effect, suggesting causality from accommodation capacity to tourism in short run.
- Intra-OECD: flights and hotel capacity less binding (small/insignificant coefficients), implying developed OECD destinations generally not capacity-constrained.

### Small island states — distinct patterns
- High tourism dependence; elasticities:
  - Origin GDP elasticities for small islands slightly smaller but within confidence intervals of non-islands.
  - Aggregate measures suggest small islands more price-sensitive, but splitting origin/destination RER shows:
    - Small islands more sensitive to origin-country real exchange movements (large significant interaction with origin PPP overvaluation).
    - Own destination RER marginal effect nearly nil: computed elasticity ≈ "0.08" (–0.124 + 0.200) with standard error "0.056" (t-stat=1.35) — short-run elasticity close to zero.
- Mechanisms:
  - Imported-heavy tourist consumption baskets and packaged vacations priced in foreign currency limit pass-through of destination depreciation to tourist-facing prices.
  - Short-run current-account improvements from depreciation likely arise via import reduction rather than increased arrivals.
- Policy-relevant note: small islands more susceptible to addition/removal of direct flights; evidence supports higher marginal flight effect (~10 percent), but reverse causality suggests caution in interpretation.

### Additional robustness and notable results
- Currency union effects:
  - Negative common-currency coefficient in main regressions driven by Eurozone; non-Euro currency unions show positive and significant coefficients.
  - Negative Eurozone result plausibly reflects many same-day visitors (Schengen) not classified as tourists by UNWTO.
- R-squared comparisons:
  - CPFE within-group R^2 suggests gravity explains higher share of variation for tourism (~20 percent) than for goods trade (~8 percent) in full sample; ranking can flip for intra-OECD flows.
- World Heritage Sites (WHS):
  - Destination WHS associated with 2–4 percent higher arrivals per WHS.
  - Origin WHS coefficient positive in full sample but negative in intra-OECD (resolving measurement/GDP correlation puzzle).
- Exchange rate measure robustness:
  - Results robust across PPP factor ratio, PPP factor misalignment (Rodrik 2008), bilateral RER, and IFS REERs; origin RER effect ~"0.27–0.30", destination RER effect ~"0.14–0.16" in controlled specs.
- Tourism receipts vs arrivals:
  - Strong correlation between log receipts and log arrivals: correlation coefficient "0.921".
  - For top decile tourism earners, first-differences regression of receipts on arrivals yields coefficient "0.869***".
  - Cross-country coefficient of variation of spending per tourist ~"0.277" (all countries), ~"0.103" for top decile tourism earners.

### Policy implications and conclusions (selected)
- Gravity model performs well for tourism and often explains a higher share of variation than for goods trade.
- Lower sensitivity of tourism to origin GDP (around "0.6") implies:
  - Relative resilience to origin-country downturns.
  - Recommendation for destinations suffering drops from traditional markets to reorient strategically to fast-growing origin markets.
- Exchange rate policy:
  - Real exchange rate movements affect tourism heterogeneously:
    - For many destinations a ten percent real depreciation implies ~1.5 percent increase in arrivals (destination elasticity ≈ "0.15").
    - Intra-OECD tourism more exchange-rate sensitive (destination elasticity around "0.35–0.4").
    - Small islands show negligible short-run tourist response to their own RER; depreciation unlikely to quickly boost arrivals.
- Connectivity and capacity:
  - Direct air connectivity and accommodation capacity matter, especially for non-OECD and small island destinations.
  - Policies to secure/subsidize air links can meaningfully affect small island tourism-dependent economies, though causality is bidirectional.
- Suggested future research (from text):
  - Use full 1995–2011 dataset to better assess dynamics, crisis-period responses, and enable dynamic GMM estimation.
  - Incorporate additional factors like visa requirements and natural disasters as data permit.

*Source: _wp1482 - Bibliography, _wp1482.pdf*

### Bibliography ...........................................................................................................

### _wp1482 - Bibliography

### Introduction and motivation
- A record one billion tourists crossed international borders in 2012, noted by the UNWTO.
- WTTC estimates travel and tourism accounts for 9 percent of global GDP from “direct and indirect activities combined”.
- In 2006–10, international tourism receipts represented about 6 percent of international trade of goods and services, and nearly 2 percent of the world’s GDP.
- International trade in fuels accounts for 10 percent of total trade; international remittances stand at ¾ percent of the world’s GDP.
- In 45 of the 191 countries and territories for which 2009 data was available, international tourism accounted for over 20 percent of total exports.
- Several OECD countries (Australia, Greece, New Zeeland, Portugal, Spain, Turkey) derive 14 to 25 percent of export earnings from foreign tourists; US, France and Italy are in the 8 to 10 percent range.
- The paper addresses a gap in the literature by analyzing the determinants of international tourism flows and cross-border spillovers using an extensive panel dataset constructed from UNWTO raw data covering the universe of bilateral tourism flows over a decade.

### Data and dataset construction
- Source: UNWTO raw data, supplemented with WDI, IFS, Penn World Tables (PWT), DOTS, and CEPII datasets.
- UNWTO electronic data distributed in five-year blocks; two blocks available for 1999–2004 and 2005–2009; as of early 2014, the full UNWTO dataset covers years 1995 through 2012.
- Key measurement heterogeneities noted:
  - Countries differ on counting at border vs. hotel, tourists vs. visitors, and origin determined by nationality vs. residence.
  - Tourist-nights collected by about 40 percent of all countries.
  - Many statistical offices group origin countries into aggregate categories (e.g., “Other Africa”, “Benelux”).
- Real exchange rate measures used:
  - (i) Bilateral real exchange rate (RER) from IFS.
  - (ii) Real effective exchange rates (REER) of origin and destination from IFS.
  - (iii) PWT’s PPP factors for origin and destination countries (bilateral PPP vis-à-vis the United States).
  - (iv) PPP factor ratio (destination PPP factor divided by origin PPP factor), interpreted as an appreciation in dollar terms when it increases.

### Dataset summary statistics (selected)
- The unbalanced panel contains 128,304 observations (each = number of tourists from origin to destination in a given year).
- Tourist arrivals dataset:
  - UNWTO full dataset: 128,304 observations; 17,441 country-pairs; 210 destinations; Total tourist arrivals, millions: 8,475; Observations kept, %: 100.0; Arrivals kept, %: 100.0
  - Country of origin unambiguously identified: 118,115 observations; 15,839 country-pairs; 204 destinations; Total tourist arrivals, millions: 7,812; Observations kept, %: 92.1; Arrivals kept, %: 92.2
  - Key variables (GDPs, distance) available: 103,676 observations; 13,573 country-pairs; 173 destinations; Total tourist arrivals, millions: 7,678; Observations kept, %: 80.8; Arrivals kept, %: 90.6
  - Minimum 100 tourists annually: 67,673 observations; 7,966 country-pairs; 173 destinations; Total tourist arrivals, millions: 7,672; Observations kept, %: 52.7; Arrivals kept, %: 90.5
- Tourist-nights dataset:
  - UNWTO full dataset: 32,152 observations; 4,208 country-pairs; 93 destinations; Total tourist-nights, millions: 9,460; Observations kept, %: 100.0; Tourist-nights kept, %: 100.0
  - Country of origin unambiguously identified: 28,315 observations; 3,685 country-pairs; 83 destinations; Total tourist-nights, millions: 7,321; Observations kept, %: 88.1; Tourist-nights kept, %: 89.0
  - Key variables (GDPs, distance) available: 24,904 observations; 3,139 country-pairs; 71 destinations; Total tourist-nights, millions: 7,064; Observations kept, %: 77.5; Tourist-nights kept, %: 87.7
  - Minimum 100 tourists annually: 19,781 observations; 2,391 country-pairs; 71 destinations; Total tourist-nights, millions: 7,062; Observations kept, %: 61.5; Tourist-nights kept, %: 87.7

### Empirical approach and scope
- Uses gravity model framework to analyze bilateral tourism arrivals and compare specifications with merchandise trade.
- Introduces a robust method for including bilateral trade flows as an independent variable in gravity estimations for non-merchandise flows.
- Examines supply-side and demand-side determinants and their impacts on arrivals and average length of stay.
- Also presents analysis for tourist-nights (smaller dataset) and a focused examination of small island destinations.

### Key empirical findings
- Fit of gravity equation:
  - The gravity equation performs very well for bilateral tourism arrivals; for some specifications, the fit for tourism is better than for goods trade.
  - Fit is considerably better than reported for FDI and remittances flows.
- Macroeconomic determinants:
  - Elasticities of bilateral tourism with respect to origin and destination GDPs are large, though smaller than near-unit elasticities estimated for goods trade.
  - Strong trade ties positively associated with higher tourism flows, indicating a considerable share of business travelers, especially intra-OECD.
  - Real exchange rate effects: appreciation of the origin’s currency increases bilateral tourism; appreciation of the destination’s currency reduces it. This elasticity is large and robust across estimation techniques and RER measures.
- Traditional gravity variables:
  - Distance between countries has a negative impact on tourism with an elasticity nearly identical to merchandise flows.
  - Language ties are more important for tourism than for merchandise trade.
  - Historical colonial relationships are less important for tourism than trade.
  - Presence of regional trade agreements associated with a small positive effect on tourism.
  - Common currency associated with a reduction in both tourism and merchandise flows; this result is fully driven by Eurozone countries.
  - Economic remoteness of the destination is associated with higher tourism flows, suggesting a premium on off-the-beaten-path destinations.
- Supply-side variables:
  - Presence of a direct flight is positively associated with tourism, but reverse causality dominates (a direct flight is added the year after bilateral tourism flows increase).
  - Number of hotel rooms in the destination is positively associated with tourism inflows; reverse causality does not seem to play a role for hotel rooms.
  - Both supply variables are less important for intra-OECD tourism flows.
- Non-traditional demand-side variables:
  - Tourists prefer regions with similar climates and have a strong preference for warmer countries.
  - Time difference has a negative impact on tourism flows, suggesting jet lag considerations.
  - Cultural capital (proxied by number of UNESCO World Heritage sites) plays a role.
  - Tourists avoid countries with ongoing conflicts.
- Tourist-nights and duration:
  - Using tourist-nights data, tourism flows respond through changes in arrivals and average duration of stay.
  - Real effective exchange rate has a strong effect on duration: real appreciation in the destination country is associated with both fewer tourists and shorter stays.
- Small islands:
  - For small island destinations, the island’s own real exchange rate has little impact on tourism arrivals, unlike other countries.

### Paper structure and supplementary materials (as described)
- Main text organized: Section II (data and summary statistics), Section III (empirical strategy), Section IV (estimation results), Section V (conclusion).
- Tables and annex tables listed include dataset summary statistics, gravity equations with various fixed effects, regressions on arrivals and tourist-nights, alternative specifications, and country coverage (Tables 1–9; Annex Tables A1–A10).

*Source: _wp1482 - Bibliography, _wp1482.pdf*

### Annex Table 2 provides totals and number of observations for each year in the dataset.

### _wp1482 - Annex Table 2 provides totals and number of observations for each year in the dataset.

### Data and key variables
- Annex Table 2 provides totals and number of observations for each year in the dataset.
- Average tourist spend: "$836 per trip."
  - Imposing a minimum of 100 tourists eliminates country-pairs with flows generating less than "US$84 thousand" in receipts for the destination country.
- CEPII’s dataset on trade agreements and currencies covers years through "2006" and was updated to reflect the expansion of the Euro area (Slovenia, Cyprus, Malta, Slovakia, Estonia).
- PPP factor ratio computed as described from PWT 7.2; identical to variable "݌" in PWT (called the “price level of GDP”) and inverse of the real exchange rate used by Rodrik (2008).
  - Interpretation: A PPP factor lower than "1" indicates the country is “cheaper” than the US; an increase indicates appreciation.
- Robustness checks also included a similar measure computed from WDI (available for a smaller subset of country-years and with weaker correlation to other exchange rate measures than the PWT-sourced measure).
- Other datasets and measures:
  - Bilateral passenger air travel flows: Diio.
  - Number of hotel beds: UNWTO.
  - World Heritage sites: UNESCO (number used is end "2012" for destination and origin WHS variable).
  - Conflicts: Political Instability Task Force (PITF) and UCDP/PRIO Armed Conflict Dataset.
  - Climate similarity index: Portland State University Koppen-Geiger classification, aggregated to seven zones (tropical rainforest, tropical savannah, steppe, desert, temperate, cold, highland). Index ranges between "0" and "1" and constructed as in Finger and Kreinin (1979).

### Empirical strategy and model specifications
- Gravity framework: tourism flows modeled analogously to bilateral merchandise trade—proportional to economic size of origin and destination and inversely related to trade resistance factors (distance, linguistic ties, historical ties, common border, trade agreements, monetary unions).
- Core panel specifications:
  - Traditional panel (no multilateral resistance controls): equation (1).
  - Country fixed effects (origin and destination dummies) to address multilateral resistance: equation (2).
    - Trade-off: impossible to estimate coefficients on time-invariant country characteristics.
  - Country-pair fixed effects (CPFE) / demeaned OLS to fully control for time-varying multilateral resistance: equation (3).
  - First-differences (FD) specification for disturbances following a random walk and to study real exchange rate impacts: equation (4).
- Approaches to estimate time-invariant variables:
  - Random effects (RE) with remoteness proxy (GDP-weighted average distance).
  - Hausman-Taylor (HT) estimator to recover time-invariant coefficients under endo-/exogeneity assumptions.
- Baseline results rely primarily on country fixed effects (CFE, equation 2) and country-pair fixed effects (CPFE, equation 3); RE and HT used to analyze additional determinants; FD used for macro determinants (real exchange rate).

### Estimation results — Tourism versus merchandise trade (country fixed effects)
- Two gravity specifications estimated: barebones (GDPs and distance) and extended (geographical, historical, linguistic controls).
- Regression sample notes:
  - Tourism regressions exclude year "2004" (missing from tourism dataset); trade regressions include "2004".
  - Discussion focuses on extended specification on full sample (tourism regression "2", trade regression "6").
- Selected coefficient estimates (Table 2, extended specifications):
  - Log Origin/Importer GDP: tourism "0.557***" (regression 2) vs. trade "0.946***" (regression 6).
  - Log Destination/Exporter GDP: tourism "1.040***" (regression 2) vs. trade "0.924***" (regression 6).
  - Log weighted distance elasticity: tourism "-1.589***" (regression 2) vs. trade "-1.541***" (regression 6).
  - Common currency: tourism "0.177***" (regression 2) vs. trade "0.177***" (regression 6) in table but discussion notes a negative coefficient for common currency in some models driven by Eurozone.
  - Members of regional trade agreement: tourism "0.252***" (regression 2) vs. trade "0.187***" (regression 6).
  - Common border: tourism "0.201***" (regression 2) vs. trade "0.427***" (regression 6).
  - Common official or primary language: tourism "0.074" (regression 2) vs. trade "0.067" (regression 6).
- Goodness of fit and observations:
  - Tourism full-sample extended specification (regression 2): Observations "104,627"; R^2 "0.848".
  - Trade full-sample extended specification (regression 6): Observations "216,038"; R^2 "0.733".
- Key qualitative findings:
  - Gravity model fits tourism and merchandise trade similarly; R^2 roughly the same and slightly better fit for tourism in comparable specifications.
  - Distance elasticities comparable: tourism "1.59" (regression 2, reported in text) vs. trade "1.54" (regression 6, reported in text).
  - Income elasticity differences:
    - Tourism arrivals elasticity with respect to origin income is much lower than goods trade importer income elasticity (robust across samples and specifications).
    - Within intra-OECD subsample, tourism income elasticity rises to just above "1" (regressions 3 and 4) but remains lower than goods trade elasticity for the same group (regressions 7 and 8).
    - Results do not support the conventional "superior good" hypothesis for tourism in the full sample.
  - Contiguity (common border) effect:
    - Common border increases number of tourists by "150 percent" for tourism (text); about "50 percent" for goods trade. (Interpretation based on exp(coef)-1.)
  - Common currency puzzle:
    - Some specifications estimate a negative coefficient for common currency for both tourism and trade; robustness checks attribute negative sign to Eurozone countries. For non-Euro currency unions, estimated coefficient varies between "0.3" and "0.6" depending on specification.

### Non-traditional determinants of international tourism (RE, CFE, HT results)
- Sample for RE/HT/CFE augmented analyses excludes country-pairs with minimum arrivals below "100" tourists in any year (reduced sample).
  - Final sample sizes: Observations "67,673"; Country-pairs "7,966" (Table 3 summary).
- Selected coefficients and statistics (Table 3 summary):
  - Log Origin GDP:
    - RE (col 1): "0.658***"
    - CFE (col 4): "0.452***"
    - HT (col 7): "0.516***"
  - Log Destination GDP:
    - RE (col 1): "0.589***"
    - CFE (col 4): "1.177***"
    - HT (col 7): "1.170***"
  - Log weighted distance:
    - RE (col 1): "-1.292***"
    - CFE (col 4): "-1.614***"
    - HT (col 7): "-1.013***"
  - Common currency:
    - RE (col 1): "0.071*"
    - CFE (col 4): "-0.088***"
    - HT (col 7): "0.028"
  - Members of regional trade agreement:
    - RE (col 1): "0.074***"
    - CFE (col 4): "0.298***"
    - HT (col 7): "0.012"
  - Common border: RE/CFE/HT all positive and large, e.g., RE "1.393***".
  - Log Origin Remoteness: RE "-0.340***"; CFE "-0.585***".
  - Log Destination Remoteness: RE "0.165***"; CFE "0.451***".
  - Log Bilateral trade (residuals): RE "0.057***"; HT "0.054***".
  - Log PPP Factor ratio (destination/origin): RE "-0.157***"; HT "-0.202***".
    - Interpretation: A "1 percent" real appreciation of the destination vis-à-vis the origin reduces arrivals by around "0.18 percent" in the CFE regression; RE gives "0.16" and HT gives "0.2" (all significant at the "1 percent" level).
  - Direct flight dummy:
    - RE "0.272***"; CFE "0.869***"; HT "0.201***".
    - Implied impact ranges from "20 percent" (HT) to "80 percent" (CFE) depending on methodology.
  - Log Destination Hotel rooms: RE "0.265***"; CFE "0.143***"; HT "0.124***".
  - Time difference in hours: RE "-0.044***"; CFE "-0.018***"; HT "-0.062***".
  - Climate Similarity Index (population-based): RE "0.367***"; CFE "0.221***"; HT "0.375***".
  - Destination conflict magnitude: RE "-0.079***"; CFE "-0.083***"; HT "-0.056***".
  - Origin World Heritage sites in 2011: RE "0.014***"; CFE "0.023***".
  - Destination World Heritage sites in 2011: RE "0.012***"; HT "-0.004" (not significant in HT).
- Goodness of fit (Table 3):
  - R^2 within examples: "0.308", "0.314", "0.317" depending on columns.
  - R^2 between examples: "0.517", "0.605", "0.681".
  - Overall R^2 examples: "0.524", "0.605", "0.675", "0.755", "0.791", "0.829" across specifications.
- Interpretation and substantive findings:
  - Gravity equation with only GDPs and distance explains over half the variance (RE) and over three-quarters (CFE) even in limited specification.
  - Population negative coefficients suggest residents of richer countries travel more and tourists prefer richer destinations.
  - Destination remoteness positive coefficient (RE and HT) suggests tourists may place a premium on destinations that are relatively far from larger economic centers ("off the beaten path").
  - Bilateral goods trade residuals enter positively and significantly, confirming economic/business ties matter for tourism (used to control for business travel).
  - Real exchange rate effects: PPP factor ratio is the preferred measure for most of the paper; a 1 percent appreciation of destination relative to origin reduces arrivals by around "0.16–0.20 percent" across methods.
  - Supply-side correlates:
    - Direct flights and hotel room capacity positively correlated with arrivals (causality not established).
    - Time zone differences reduce tourism (jet lag effect).
  - Cultural capital (UNESCO WHSs):
    - Origin WHS count positively correlated with residents traveling abroad (robust).
    - Destination WHS effect mixed: RE shows each WHS associated with "1 percent" higher arrivals; HT finds no statistically significant effect.
  - Climate similarity:
    - Positive and significant correlation — tourists prefer similar climates rather than different climates (familiarity effect dominates).
  - Conflict:
    - Destination conflict magnitude reduces arrivals (negative and significant coefficients).

### Robustness and methodological notes
- Exclusion threshold: reduced sample excludes country-pairs with flows below "100" tourists in any year.
- Hausman and Arellano tests reject random effects specification, yet RE results presented as comparator for HT.
- PPP factor ratio preferred over REER for most of the paper because REER weights merchandise trade heavily and may be less relevant for tourism-dependent economies; PWT-sourced measures benchmark country-pairs to a single reference country and conceptually lie between bilateral RER and REER.
- Robustness checks used WDI-sourced PPP factors (smaller coverage, weaker correlations) and alternative specifications; detailed robustness results available upon request.

*Source: _wp1482 - Annex Table 2 provides totals and number of observations for each year in the dataset.*

### annex tables. Table A5 features random effects results for the full sample (not restricted to

### _wp1482 - annex tables. Table A5 features random effects results for the full sample (not restricted to

### Overview
- Annex tables include Table A5 (random effects results for the full sample, not restricted to country-pairs with over 100 tourists per year), Table A6 (intra-OECD tourism), and Table A7 (alternative specifications).
- Results compare baseline regressions to alternative samples and specifications.

### Full sample (Table A5, regressions 9 through 11)
- Results for the full sample (Table A5 regressions 9 through 11) are broadly in line with baseline results.
- The coefficient on GDPs is closer to one.
- The magnitude and significance of other key variables are broadly unchanged.

### Intra-OECD sample (Table A6)
- Economic variables have much larger effects in the intra-OECD sample (Table A6):
  - GDP elasticities are close to one.
  - The impact of the real exchange rate jumps to nearly 0.3 (from less than 0.2 estimated for the world).
  - The coefficient on bilateral trade is between 0.13 and

### Alternative specifications (Table A7)
- Table A7 looks at alternative specifications. (No further detail provided in the supplied content.)

*Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1482.pdf*

### 0.3 across specifications (up from the 0.05 to 0.2 range for the world). Results for intra-

### _wp1482 - 0.3 across specifications (up from the 0.05 to 0.2 range for the world). Results for intra-

### Major empirical approach and data
- Method: gravity model applied to bilateral tourism flows (tourist arrivals and tourist-nights) using multiple estimators: Random Effects, Country-pair Fixed Effects (CPFE), First Differences (FD), Hausman-Taylor, CFE.
- Data sources and scope: UNWTO bilateral arrivals and nights, PWT 7.2 PPP factors, IFS REERs, CEPII geography variables; sample occasionally restricted to pairs with minimum 100 tourists per year; sample windows referenced include 1999–2009 with additional datasets up to 2011 for country-level receipts/nights.
- Model controls include: origin and destination GDPs and populations, weighted distance, bilateral trade residuals, direct flight dummy, destination hotel rooms, conflict magnitude, climate similarity, World Heritage Sites, common currency / Eurozone / non-Euro currency dummies, RTA dummies, year fixed effects, and various interaction terms (e.g., island × exchange rate, border × origin GDP).

### Key findings: income elasticities and gravity variables
- Origin GDP elasticity for tourism arrivals:
  - Full sample: around 0.6 (e.g., 0.580***, 0.610*** in CPFE and FD; 0.658***, 0.701*** in Random Effects specifications appear).
  - Intra-OECD sample: elasticity significantly higher—often above 1 (examples: 1.263***, 1.190*** in CPFE/FD; Table 6 intra-OECD regressions report origin GDP in the 0.94–1.48 range across specifications).
  - Conclusion: tourism exhibits a lower elasticity with respect to origin GDP than goods trade (tourism ~0.6 vs goods trade ~1 in many specifications); intra-OECD tourism elasticity climbs back to around one.
- Destination GDP elasticity:
  - Larger than origin GDP elasticity for tourism in most specifications (examples: Log Destination GDP 1.173***, 1.050*** in CPFE; 0.804***, 0.694*** in FD depending on controls).
- Bilateral trade residuals:
  - Positive and significant association with tourism; e.g., Log Bilateral trade (residuals) coefficients: 0.050*** (Random Effects/HT), 0.012***–0.036*** across CPFE/FD and nights regressions — stronger correlation for intra-OECD flows (suggesting business travel link).
- Distance and geography:
  - Weighted distance robustly negative (e.g., -1.292***, -1.026*** in Random Effects/H-T; -1.023*** in Table A7).
  - Common border large positive effect (e.g., around 1.3–1.4 in many specifications).
  - Climate similarity positive (e.g., 0.367***; tourists prefer traveling to countries with similar climates).
- World Heritage Sites (WHS):
  - Destination WHS: associated with 2–4 percent higher arrivals to the destination (text: “each WHS is associated with 2–4 percent higher arrivals”).
  - Origin WHS coefficient can be positive in full sample but this likely reflects correlation with origin GDP and measurement error; OECD sample helps resolve puzzle (origin WHSs coefficient negative in intra-OECD).

### Real exchange rate and price sensitivity
- Strong correlation between bilateral real exchange rate and tourism arrivals across multiple RER measures.
- Bilateral RER elasticity:
  - Around -0.2 (text: “elasticity of tourism arrivals with respect to bilateral exchange rate is around 0.2”; Table 6: Log Bilateral real exchange rate -0.186*** to -0.213*** across specs).
- Destination vs origin RER split (first-difference specifications):
  - Origin’s real exchange rate (PPP factor / REER) enters positively and larger in magnitude: examples include Log Origin PPP factor 0.266*** (regression 2), Log Origin REER ~0.274***.
  - Destination’s real exchange rate enters negatively but smaller in magnitude: examples include Log Destination PPP factor -0.084*** (regression 2), Log Destination REER ~-0.106*** to -0.156***.
  - After additional controls, elasticities reported: origin’s REER ≈ 0.27–0.30, destination’s REER ≈ 0.14–0.15 (text: “elasticity of tourism arrivals with respect to the origin’s real exchange rate is close to 0.3, while the elasticity with respect to the destination’s real exchange rate is around 0.14–0.15”).
- Interpretation and causality:
  - Reverse causality primarily biases destination RER coefficient downwards; the origin-country coefficient (~0.28) may provide an upper bound for the true destination effect.
  - For intra-OECD tourism, RER elasticities are about twice higher than for the full sample; destination RER elasticity for intra-OECD reported around 0.35–0.4 after controls.
- Magnitude implication: a ten percent real depreciation of the destination is associated with a 1.5 percent increase in tourism arrivals when using the estimated destination elasticity ≈ 0.15.

### Tourist-nights, average stay, and extensive vs intensive margins
- Tourist-nights are conceptually closer to tourism revenues; regressions reveal:
  - Tourist-nights more sensitive to destination real exchange rate movements than arrivals.
  - Intensive margin (change in nights / duration) measured as difference between nights and arrivals coefficients: example given — intensive margin on the destination PPP misalignment ≈ 0.41 (0.744–0.333).
- Average tourist stay regressions (nights / arrivals) findings:
  - Intensive margin with respect to origin GDP and origin real exchange rate is nil: richer origin countries and origin appreciation increase travel incidence but not length of stay.
  - Tourists reduce length of stay when the destination real exchange rate appreciates (i.e., higher destination prices lead to shorter stays).
  - As destination countries grow, average duration of stay goes down (linked to greater share of business travel).
  - Stronger trade connections associated with shorter stays (likely business travel).
- Combined effect: some specifications imply overall impact of a 1 percent change in destination real exchange rate on tourism flows exceeds 0.7 when measured in tourist-nights; bilateral RER-focused regression reports elasticity of 0.45 (Table 8 discussion).

### Supply-side variables: direct flights and hotel capacity
- Direct flight:
  - CPFE (full sample) shows presence of a direct flight associated with 22 percent higher bilateral tourism flows (computed as exp(0.2)-1 using CPFE coefficient ~0.2).
  - FD estimates the contemporaneous addition of a flight is associated with a 4 percent increase in tourism (direct flight coefficient 0.040*** in FD; text: “FD reports that the addition of a flight is associated with a 4 percent increase in tourism”).
  - Both lag and lead of direct flight are highly significant; forward coefficient larger in magnitude — evidence that flights are often established in response to rising tourism (causality goes both ways; net implication: addition of a direct flight follows increase in bilateral tourism).
  - Small islands: addition/removal of a direct flight has stronger marginal effect — interaction implies marginal effect of 0.1 (10 percent) with s.e. 0.03 (significant at 1 percent) for small island destinations.
- Hotel rooms:
  - CPFE and FD show positive and significant associations: Log Destination Hotel rooms coefficients e.g., 0.138***, 0.246*** in various specifications.
  - FD coefficient often twice larger than CPFE, suggesting stronger short-run effect from adding hotel rooms (e.g., opening promotions), and that causality runs from accommodation capacity to tourism rather than the reverse.
  - For intra-OECD pairs, flights and hotel capacity are less binding (coefficients small/insignificant), implying developed OECD destinations generally not capacity-constrained.

### Small island states: special considerations
- Small islands receive high shares of GDP from tourism and are highly tourism-dependent.
- Elasticity patterns:
  - Origin GDP elasticities for small islands slightly smaller but within confidence intervals of non-islands.
  - Measures that do not split origin/destination RER suggest small islands appear more price-sensitive; deeper analysis shows:
    - Small islands are more sensitive to origin-country real exchange movements (large significant interaction with origin PPP overvaluation).
    - For their own destination RER, marginal effect is nearly nil: computed elasticity ≈ 0.08 (–0.124 + 0.200) with standard error 0.056 (t-stat=1.35) — i.e., short-run elasticity close to zero.
- Mechanisms:
  - Imported-heavy tourist consumption baskets and packaged vacations (prices set/negotiated in foreign currency) limit the pass-through of destination currency depreciation to tourist-facing prices.
  - Short-run current-account improvements from depreciation likely arise via import side reduction, not increased tourist arrivals.
- Supply sensitivity: small islands more susceptible to addition/removal of direct flights; evidence supports higher marginal effect of direct flight for small islands (~10 percent), although reverse causality suggests flight → tourism effect may be smaller than 10 percent.

### Additional notable empirical results and robustness
- Currency union effects:
  - Negative common-currency coefficient in main regressions is driven by the Eurozone; non-Eurozone common-currency areas show positive and significant coefficients (Table A7 regressions decomposing currency dummy).
  - Negative Eurozone result plausibly explained by higher prevalence of same-day visitors in Schengen, who are not classified as tourists by UNWTO but play similar economic role.
- R-squared comparisons:
  - CPFE within-group R2 suggests gravity equation explains a higher share of variation for tourism (~20 percent) than for goods trade (~8 percent) in full sample; ranking flips for intra-OECD flows.
- World Heritage Sites (WHS) (OECD-specific):
  - Each WHS associated with 2–4 percent higher arrivals to the destination; origin WHS coefficient negative in intra-OECD (resolving full-sample puzzling positive origin WHS coefficient due to correlation with GDP and measurement error in poorer countries).
- Robustness of exchange rate measures:
  - Results robust across PPP factor ratio, PPP factor misalignment (Rodrik 2008 adjustment), bilateral real exchange rate, and IFS REERs; origin’s RER effect consistently ~0.27–0.30 while destination’s RER effect ~0.14–0.16 in fully controlled specifications.
- Tourism receipts vs arrivals:
  - Strong correlation between log tourism receipts and log tourist arrivals: correlation coefficient 0.921.
  - For top decile tourism earners, first-differences regression of receipts on arrivals yields coefficient 0.869*** (suggesting arrivals are a good predictor of receipts for large tourism economies).
  - Cross-country variation in spending per tourist exists; coefficient of variation ~0.277 (all countries), lower (~0.103) for top decile tourism earners.

### Policy implications and conclusions (selected)
- Gravity model performs well for tourism and often explains a higher share of variation than for goods trade.
- Tourism’s lower sensitivity to origin GDP (around 0.6) implies:
  - Relative resilience to origin-country downturns (can be beneficial if traditional markets slump).
  - Strategic reorientation to fast-growing origin markets (e.g., China) is a recommended policy response for destinations suffering drops from traditional markets.
- Exchange rate policy:
  - Real exchange rate movements affect tourism, but the magnitude and sign differ across country types:
    - For many destinations a ten percent real depreciation implies ~1.5 percent increase in arrivals (destination elasticity ≈ 0.15).
    - Intra-OECD tourism is more exchange-rate sensitive (destination elasticity around 0.35–0.4).
    - Small islands show negligible short-run tourist response to their own RER, so depreciation is unlikely to quickly boost arrivals; short-term external adjustment likely operates via import side.
- Supply-side and connectivity:
  - Direct air connectivity and accommodation capacity matter, especially for non-OECD and small island destinations.
  - Policies to secure and subsidize air links can have meaningful impacts for small island tourist-dependent economies, although causality is bidirectional.
- Future research suggestions (from text):
  - Use full 1995–2011 dataset (paper used a subset with 2004 missing) to better assess dynamics, crisis-period responses, and to enable improved dynamic GMM estimation.
  - Incorporate additional factors like visa requirements and natural disasters as data permit.

*Source: IMF working paper content provided in the input PDF unit.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1482.pdf_