BƯỚC 4. ENDOGENEITY, IDENTIFICATION VÀ ADVANCED ANALYSES

Bài mẫu: Tourism Uncertainty and Corporate Dividend Policy

Mục tiêu: phân biệt một association giữa TUI và payout với một design đủ mạnh để nói TUI thực sự liên quan đến quyết định dividend policy qua các channels kinh tế cụ thể.

PHẦN A. ENDOGENEITY DIAGNOSIS

Concern	Severity	Why
Omitted country shocks	High	tourism conditions, policy, macro cycle affect both TUI and dividends
Generic uncertainty	High	TUI may proxy EPU/GPR/WUI
Tourism demand shock	High	weaker demand can raise TUI and reduce earnings/dividends
Reverse causality	Low at firm level	one firm’s dividend unlikely to move national TUI
Dynamic payout persistence	High	dividends are sticky
Country-level treatment	High	inference depends on country shocks
Tourism-firm selection	Medium	tourism firms structurally differ from non-tourism firms

The central problem is country-year omitted shocks, not firm-level reverse causality.

PHẦN B. STRONGEST DESIGN — TOURISM EXPOSURE INTERACTION

Use all firms in TUI countries and estimate:

$$
DividendPolicy_{i,c,t+1}

beta
(TUI_{c,t}times TourismExposure_i)
+
FirmFE_i
+
Countrytimes YearFE
+
Industrytimes YearFE
+
varepsilon
$$

Country×Year FE absorb:

• TUI main effect;
• GDP;
• inflation;
• country tourism demand;
• country policy;
• COVID-country effects;
• all other country-year shocks.

Identification asks:

Within the same country-year, do tourism-exposed firms adjust payouts differently from non-tourism firms?

This is a very strong design if TourismExposure is predetermined.

PHẦN C. TOURISM EXPOSURE MEASURES

1. Industry Dummy

Simple:

$$
TourismExposure_i=1
$$

for tourism industries.

2. Tourism Dependence Intensity

If firm/business-segment data exist:

• tourism revenue share;
• hotel/airline/travel segment share;
• leisure share.

3. Country Tourism Dependence

Alternative:

$$
TourismReceipts/GDP
$$

or tourism employment share.

Then:

$$
TUI_{ct}times CountryTourismDependence_c
$$

Caution

This is country-year if dependence varies with time and may be absorbed by Country×Year FE.

Prefer:

• pre-sample;
• pre-2019;
• lagged/predetermined dependence.

PHẦN D. PREDETERMINED EXPOSURE

Use:

$$
Exposure^{Pre}
$$

measured before the main TUI shock/sample period where possible.

Why?

Current tourism exposure may itself react to uncertainty.

PHẦN E. TOURISM DEMAND CHANNEL

Potential chain:

$$
TUIuparrow
rightarrow TourismDemanddownarrow
rightarrow CashFlowdownarrow
rightarrow Dividendsdownarrow
$$

Evidence can come from:

• tourism revenue growth;
• sales growth;
• operating cash flow;
• profitability;
• occupancy/RevPAR if external data available.

If only Compustat:

use Sales Growth/CFO/Profitability as channel outcomes.

Do not automatically call them mediators without proper identification.

PHẦN F. PRECAUTIONARY CASH CHANNEL

Test whether TUI predicts:

$$
CashHoldingsuparrow
$$

and:

$$
Dividendsdownarrow
$$

A combined pattern supports liquidity preservation.

Potential stronger test:

$$
TUItimes LowCash
$$

or:

$$
TUItimes FinancialConstraints
$$

If constrained firms cut more:

strong precautionary interpretation.

PHẦN G. SIGNALING / REASSURANCE CHANNEL

If TUI increases dividends:

look for:

• stronger effect in firms with information asymmetry;
• weak investor protection;
• high uncertainty;
• stable profitability;
• mature firms.

Interpretation:

firms use dividends to reassure investors.

Do not claim signaling just because β > 0.

PHẦN H. AGENCY CHANNEL

If TUI increases payout:

test whether effect is stronger among firms with:

• high cash;
• high free cash flow;
• weak governance;
• low growth opportunities.

This would be more consistent with agency-cost mitigation.

PHẦN I. DIVIDEND SMOOTHING / PARTIAL ADJUSTMENT

Dynamic model:

$$
Dividend_{it}

alpha
+
rho Dividend{i,t-1}
+
beta TUI{ct}
+
…
$$

Possible implication:

• high ρ = sticky dividends;
• uncertainty affects target payout more than immediate payout.

Caution

Lagged Y + FE causes dynamic bias in short panels.

If using System GMM:

report all diagnostics.

PHẦN J. CUT / OMISSION IDENTIFICATION

Because firms dislike cuts:

TUI may predict:

• lower dividend growth;
• higher cut probability;
• omission only under severe uncertainty.

Use:

$$
HighTUI
$$

or nonlinear specification.

Potential threshold:

$$
TUI > P75
$$

Do not create thresholds only after seeing data without disclosure.

PHẦN K. COVID AS QUASI-STRESS EPISODE

COVID is an extreme tourism shock.

Possible interaction:

$$
TUItimes COVID
$$

But COVID is not a clean instrument for TUI.

It directly affects:

• earnings;
• liquidity;
• operations.

Use as heterogeneity/stress test, not IV.

PHẦN L. IV SCREENING

Poor automatic instruments:

• tourist arrivals;
• GPR;
• EPU;
• COVID;
• exchange rates;
• oil prices.

Each can directly affect dividend policy.

A credible IV would need:

1. strong relation to TUI;
2. no direct payout effect except through TUI.

This is difficult.

Prefer exposure/FEs to weak IV.

PHẦN M. HETEROGENEITY PRIORITY

1. tourism-exposed vs non-tourism firms;
2. airlines / hotels / restaurants;
3. financial constraints;
4. cash holdings;
5. leverage;
6. retained earnings / life-cycle;
7. profitability;
8. country tourism dependence;
9. financial development;
10. crisis periods.

PHẦN N. MARGINAL EFFECTS

For:

$$
Dividend

beta_1TUI+beta_2M+beta_3TUItimes M
$$

report:

$$
frac{partial Dividend}{partial TUI}

beta_1+beta_3M
$$

at:

• P25;
• P50;
• P75.

PHẦN O. ADVANCED PACKAGE RECOMMENDED

A strong paper can use:

1. country TUI baseline with Firm + Year FE;
2. fiscal-year TUI;
3. general uncertainty controls;
4. Country×Year FE exposure interaction;
5. tourism-only vs non-tourism comparison;
6. cash-flow channel;
7. cash-preservation channel;
8. constraints interaction;
9. dividend cuts/omissions;
10. dynamic smoothing model;
11. COVID stress heterogeneity.

PHẦN P. SAMPLE WRITING

The principal identification concern is that country-level Tourism Uncertainty may co-move with macroeconomic and tourism-demand conditions that also affect firms’ payouts. We therefore complement the direct TUI specification with a tourism-exposure design. By interacting TUI with predetermined tourism exposure and including country-by-year fixed effects, we absorb the main TUI series together with all other country-year shocks. The remaining coefficient identifies whether tourism-exposed firms adjust payout policy differently from otherwise comparable firms facing the same country-year environment. We further examine cash-flow, liquidity, financial-constraint, and dividend-smoothing channels to distinguish precautionary payout behavior from signaling or agency explanations.

PHẦN Q. OUTPUT

1. Endogeneity Diagnosis Table.
2. Tourism Exposure Interaction Table.
3. Country Tourism Dependence Table.
4. Cash-Flow Channel Table.
5. Cash-Preservation Table.
6. Signaling/Agency Table.
7. Dividend Cut/Omission Table.
8. Dynamic Smoothing Diagnostics.
9. COVID Heterogeneity.
10. Marginal Effects.
11. Advanced analyses draft 1,800–2,800 words.
12. Final causal-language decision.

PHẦN R. CHECKLIST

☐ main omitted-shock problem stated.
☐ tourism exposure design implemented if feasible.
☐ Country×Year used only after main TUI is absorbed.
☐ exposure predetermined where possible.
☐ tourism demand channel tested.
☐ cash-preservation channel tested.
☐ signaling not inferred from sign alone.
☐ agency channel separately tested.
☐ dynamic smoothing handled carefully.
☐ COVID not used as naive IV.
☐ IV excluded if exclusion restriction weak.
☐ marginal effects reported.

Chuyển sang Bước 5: Theory, Literature Review & Hypotheses.