## Introduction

## Source details

**Canonical URL:** [Introduction](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023153-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023153-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023153-print-pdf.pdf.json)

---

### Purpose and problem
- Nowcasting uses higher frequency data to produce forecasts of lower frequency indicators, providing timelier information for imminent policy changes.
- Standard nowcasting tools are reduced-form and cannot assess the impact of proposed policy changes on baseline nowcasts; the paper proposes two methodologies to address this shortcoming.
- The application yields estimates of the sensitivity of a nowcast to changes in chosen policy indicators at least one period ahead.
- Assumptions:
  - A single-equation regression nowcasts real GDP using lags of real GDP and at least one high-frequency indicator converted to the lower frequency of real GDP.
  - Not enough observations exist to estimate a structural VAR at the frequency of real GDP; feasible SVARs would be poorly specified.

### Major themes and structure
- Introduction: page 5
- First Approach: page 6
- First Example: Limited Data, No Feedback Effects: page 7
- Second Approach: page 12
- Second Example: Limited Data, General Feedback Effects: page 14
- Conclusion: page 16
- References: page 16

### First approach: single-equation ARMA(p, q) models (and possible VAR extensions)

### Method overview
- For each high-frequency explanatory variable in the baseline model, estimate a single-equation ARMA(p,q) with an exogenous policy instrument.
- Use the preferred ARMA(p,q) to generate a conditional forecast for the high-frequency variable, convert to the target lower frequency (e.g., sum, averaging).
- Insert the high-frequency forecast into the baseline regression for real GDP (also including the policy instrument exogenously) to obtain a nowcast reflecting the proposed policy change.
- The approach provides a one-period ahead estimate unless there is no feedback from real GDP to the policy instrument.

### Extensions
- High-frequency indicators can be forecasted jointly via multi-equation structural models (VARs or SVARs) including an exogenous proxy for policy instruments; convert forecasts to target frequency and compare revised nowcast to baseline.

### Limitations
- The methodology is necessarily one-period ahead when feedback effects from the target variable to policy instruments or other RHS variables are present.
- If used for more than one period, the ARMA(p,q) must not be re-estimated after imposing the shock, otherwise impacts will be affected by parameter changes.

### First example: Dominica — limited data, no feedback effects

### Context and baseline regression
- Small open economy where external factors (USA real GDP, real WTI crude oil price) are treated as exogenous to Dominica.
- Baseline annual nowcast for Dominica’s real GDP uses four higher-frequency indicators: real domestic imports (dm_im, monthly), tourism expenditures (dm_vis_e, monthly), primary balance (dm_pb, monthly), return on average equity (dm_roe, quarterly); plus two exogenous variables rgdp_usa (quarterly) and r_wti (monthly).
- Baseline regression (annual, 2001-2021): Dependent variable DLOG(RGDP)
  - Sample: 2001 2021; Included observations: 21
  - Coefficients:
    - C — -0.023209 — 0.010849 — -2.139346 — 0.0505
    - DLOG(DM_IM) — 0.181250 — 0.046806 — 3.872325 — 0.0017
    - DLOG(DM_VIS_E) — 0.051998 — 0.017337 — 2.999274 — 0.0096
    - D(DM_PB) — 0.000422 — 0.000249 — 1.694686 — 0.1123
    - DM_ROE — 0.000749 — 0.000310 — 2.419824 — 0.0297
    - D(R_WTI) — -0.014054 — 0.009818 — -1.431469 — 0.1742
    - D(RGDP_USA) — 0.009849 — 0.003708 — 2.656161 — 0.0188
  - Goodness-of-fit and diagnostics:
    - R-squared 0.881552
    - Adjusted R-squared 0.830788
    - S.E. of regression 0.023466
    - F-statistic 17.36583; Prob(F-statistic) 0.000009
    - Durbin-Watson stat 2.288516

### Example ARMA model for dm_vis_e
- Quarterly, 2000Q2-2021Q4; preferred ARMA(2,1) with CRISIS dummy, DLOG(RGDP_USA), R_WTI
  - Sample: 2000Q2 2021Q4; Included observations: 87
  - Coefficients:
    - C — 51.36789 — 10.46747 — 4.907385 — 0.0000
    - CRISIS — -7.346555 — 4.069085 — -1.805456 — 0.0748
    - DLOG(RGDP_USA) — 115.8278 — 37.74850 — 3.068408 — 0.0029
    - R_WTI — -2.209432 — 5.524214 — -0.399954 — 0.6903
    - AR(1) — 1.459534 — 0.487461 — 2.994153 — 0.0037
    - AR(2) — -0.536291 — 0.447560 — -1.198255 — 0.2344
    - MA(1) — -0.412037 — 0.561082 — -0.734362 — 0.4649
    - SIGMASQ — 79.11120 — 9.889257 — 7.999711 — 0.0000
  - Model fit:
    - R-squared 0.833063; Adjusted R-squared 0.818271; S.E. of regression 9.333941
    - Akaike info criterion 7.412926; Schwarz criterion 7.639676; Durbin-Watson stat 1.994653

### Steps to produce nowcast and scenarios
- Convert monthly data to quarterly to match rgdp_usa; extend quarterly forecasts to 2022Q4 using ARMA(p,q) models, then convert to annual frequency by averaging quarterly observations.
- Use extended annual high-frequency series in baseline regression estimated with 2001-2021 to nowcast 2022 RGDP.
- Create alternative scenarios for exogenous external factors (e.g., 10 percent increase in real oil prices for 2022; 1 percent increase in USA real GDP for 2022), recalculate high-frequency forecasts where affected, recompute nowcast, and compare to baseline.

### Dominica scenario results (one-period ahead impacts)
- Baseline real GDP nowcast for 2022: 9.55 percent year-on-year growth (or 1217.62 million ECU dollars).
- Given a 10 percent increase in real oil prices for 2022:
  - Revised nowcast: 9.43 percent (or 1216.35 million ECU dollars).
  - Interpretation: one-period ahead impact of higher oil prices is rather small.
- Given a 1 percent increase in USA real GDP during 2022:
  - Revised nowcast: 11.34 percent year-on-year growth (or 1237.51 ECU dollars) compared to baseline 9.55 percent.

### Multi-period extension
- Because USA real GDP and global oil prices are exogenous to Dominica, the baseline regression can provide multi-period estimates; high-frequency variables must be extended (e.g., to 2023) while keeping baseline parameter estimates unchanged.

### Second approach: non-parametric structural estimator (Ouliaris and Pagan, 2022)

### Method overview
- A non-parametric structural estimator estimates responsiveness of a nowcast to a structural shock without requiring full identifying restrictions.
- Procedure: simulate structural parameters for the structural equation of interest to estimate the structural shock, then regress other variables on the estimated shock from a reduced-form VAR to obtain responses.
- Advantages in nowcasting context:
  - Low number of parameters relative to VAR.
  - Appropriately agnostic about identifying restrictions underlying true impulse responses.

### Triangular SVAR illustration (three-variable system)
- Structural equation for high-frequency variable x1t:
  - x1t = α12 x2t + α13 y3t + ... + η1t
- Reduced-form equations:
  - x2t = ... + ρ1 η1t + v2t
  - y3t = ... + ρ2 η1t + v3t
- Estimation:
  - Simulate α12 and α13 consistent with economic priors; obtain residuals η̂1t.
  - Regress x2t and y3t on lags of endogenous variables and η̂1t to estimate ρ1 and ρ2 by OLS; ρ2 is the response of y3t to a unit shock in x1t.
  - If x1t is exogenous, α12 and α13 = 0, and η1t can be replaced by x1t.

### Second example: Tonga — limited data, general feedback effects

### Context and baseline regression
- Tonga: small Polynesian economy with frequent natural shocks and heavy reliance on remittances; main partners Australia, New Zealand, USA.
- Annual real GDP released with 1-2 year lag; semi-annual nowcasting model estimated.
- Baseline semi-annual regression for Tonga (2011S2-2021S1): Dependent variable DLOG(GDPR_CH)
  - Sample (adjusted): 2011S2 2021S1; Included observations: 20
  - Coefficients:
    - C — -0.000666 — 0.006342 — -0.105059 — 0.9186
    - DLOG(GDPR_CH(-1)) — 0.447579 — 0.221554 — 2.020177 — 0.0741
    - DLOG(TOTAL_AGR) — 0.002912 — 0.006289 — 0.462935 — 0.6544
    - DLOG(CONST_PERR(-1)) — 0.015971 — 0.007837 — 2.037828 — 0.0720
    - DLOG(CONST_PERR(-2)) — 0.011338 — 0.008927 — 1.270036 — 0.2359
    - DLOG(REM_R) — 0.009365 — 0.017558 — 0.533381 — 0.6067
    - DLOG(REM_R(-1)) — 0.004907 — 0.015061 — 0.325794 — 0.7520
    - CRD_GRW(-1) — 0.000128 — 0.000295 — 0.433359 — 0.6750
    - DLOG(TRV_RECR) — 0.005767 — 0.017392 — 0.331609 — 0.7478
    - DLOG(AUS_GDP(-1)) — 0.191812 — 0.424280 — 0.452088 — 0.6619
    - CRISIS — -0.003879 — 0.010082 — -0.384716 — 0.7094
  - Fit and diagnostics:
    - R-squared 0.728796; Adjusted R-squared 0.427458; S.E. of regression 0.009060
    - Akaike info criterion -6.268467; Prob(F-statistic) 0.099773; Durbin-Watson stat 1.325430

### Problem and applicability of non-parametric approach
- Large number of explanatory variables with limited observations (20 semi-annual observations) makes conventional structural approaches infeasible.
- The non-parametric sign-restricted estimator can be used to assess the impact of shocks (e.g., to travel receipts) on the nowcast without estimating a full structural model.

### Application to tourism shock and estimation procedure
- Structural equation for travel receipts (trv_recr) specified as function of current real GDP (gdpr_ch), remittances (rem_r), total agricultural production (total_agr) plus lagged terms and constant; structural error η1t captures the shock to travel receipts.
- Procedure:
  - Simulate β1, β2, β3 from a random distribution (e.g., uniform), estimate residuals from the structural equation to obtain η̂1t for each draw.
  - Use reduced-form VAR equation for real GDP with η̂1t as regressor to estimate ρ2 by OLS; ρ2 is the response of the nowcast of real GDP to a positive unit shock in travel receipts.
  - Search over feasible space for β1, β2, β3 and compute summary statistics for ρ2 (mean, median, max, min) to obtain robust estimates for adjusting baseline nowcast given assumed shock size.
- Implementation choices in the Tonga example:
  - Regression in equation (4) uses data regressed on a constant and crisis dummy.
  - To avoid unintended constraints, assumed range (−∞, ∞) for β1, β2, β3.
  - Random values di for β1, β2, β3 were drawn from uniform (−1, 1) and transformed to (−∞, ∞) using 1/(1−lal(di)). One million estimates of η1t and ρ2 were generated.

### Empirical result (tourism shock sensitivity)
- The estimates for ρ2 from equation (5) suggest that, on average, there is a positive relationship between the tourism sector and real GDP.
- Key statistical sensitivity estimate:
  - The sensitivity of the baseline nowcast for real GDP to a unit shock arising from the tourism sector (i.e., travel receipts) one period ahead is summarized by: 0.56 percent, a median of 0.24 percent and an interquartile range of [0, 1.12].
  - These statistics can be used to estimate the sensitivity of the baseline nowcast for real GDP to a unit shock from the tourism sector one period ahead.

### Methodologies proposed and practical considerations

### Methodologies
- Partial equilibrium approach:
  - Based on an AAAAAA(p,p?) model as presented in the paper (notation in source: 퐴퐴퐴퐴퐴퐴퐴퐴(푝푝,푞푞)).
  - Regression-based and produces one period estimates of the impact of a policy change.
- Non-parametric structural estimator:
  - Based on the sign-restricted methodology.
  - Non-parametric and produces one period estimates of the impact of a policy change.
- The two approaches:
  - Can be shown to be theoretically equivalent when estimating the impact of exogenous variables.
  - Both yield one period estimates useful for policy analysis.

### Practical considerations for implementation
- Work with residuals after regressing on a constant and the crisis dummy (residuals of 푑푑푙푙푑푑푙푙(푡푡푡푡푣푣_푡푡푟푟푟푟푡푡/푡푡) as specified).
- All required variables in the regression are included with a single lag.
- For the non-parametric sign restrictions, the endpoints of the range (i.e., −∞ and ∞) are excluded.

### Conclusion and policy relevance
- The sensitivity of the baseline nowcast for a low frequency variable from a structural shock to a high frequency explanatory variable is critical information for policymakers.
- The paper provides two regression-based tools that allow policymakers to estimate the one period impact of policy changes on nowcasts, illustrated with country-specific examples (Dominica and Tonga).

*Source: IMF Working Paper No. WP/2023/153 — Assessing the Impact of Policy Changes on a Nowcast*

### Introduction ...........................................................................................................

### Introduction

### Major themes and structure
- Introduction: page 5
- First Approach: page 6
- First Example: Limited Data, No Feedback Effects: page 7
- Second Approach: page 12
- Second Example: Limited Data, General Feedback Effects: page 14
- Conclusion: page 16
- References: page 16

### Figures listed in the content unit
- Figure 1: Year-on-Year Percentage Change, 2019-2022, Real GDP of Dominica
- Figure 2: Year-on-Year Percentage Change, 2019-2023, Real GDP of Dominica

### Tables listed in the content unit
- Table 1: Baseline Nowcasting Regression for the Real GDP of Dominica, 2001-2021(Annual)
- Table 2: Preferred ARMA(p, q) Model for Tourist Expenditures in Dominica, 2000Q2-2021Q4 (Quarterly)
- Table 3: Annual Data for Baseline Nowcasting Regression
- Table 4: Year-On-Year Percentage Change, 2019-2022, Real GDP of Dominica
- Table 5: Year-On-Year Percentage Change, 2019-2023, Real GDP of Dominica
- Table 6: Baseline Nowcasting Regression for Tonga, 2011S1-2021S1

*IMF WORKING PAPERS Assessing the Impact of Policy Changes on a Nowcast — INTERNATIONAL MONETARY FUND*

### Introduction

### wpiea2023153-print-pdf - Introduction

### Introduction: purpose and problem
- Nowcasting uses higher frequency data to produce forecasts of lower frequency indicators, providing timelier information for imminent policy changes.
- Standard nowcasting tools are reduced-form and cannot assess the impact of proposed policy changes on baseline nowcasts; the paper proposes two methodologies to address this shortcoming.
- The application yields estimates of the sensitivity of a nowcast to changes in chosen policy indicators at least one period ahead.
- Assumptions:
  - A single-equation regression nowcasts real GDP using lags of real GDP and at least one high-frequency indicator converted to the lower frequency of real GDP.
  - Not enough observations exist to estimate a structural VAR at the frequency of real GDP; feasible SVARs would be poorly specified.

### First approach: single-equation ARMA( p, q ) models (and possible VAR extensions)
- Method overview:
  - For each high-frequency explanatory variable in the baseline model, estimate a single-equation ARMA(p,q) with an exogenous policy instrument.
  - Use the preferred ARMA(p,q) to generate a conditional forecast for the high-frequency variable, convert to the target lower frequency (e.g., sum, averaging).
  - Insert the high-frequency forecast into the baseline regression for real GDP (also including the policy instrument exogenously) to obtain a nowcast reflecting the proposed policy change.
  - The approach provides a one-period ahead estimate unless there is no feedback from real GDP to the policy instrument.
- Extensions:
  - High-frequency indicators can be forecasted jointly via multi-equation structural models (VARs or SVARs) including an exogenous proxy for policy instruments; convert forecasts to target frequency and compare revised nowcast to baseline.
- Limitations:
  - The methodology is necessarily one-period ahead when feedback effects from the target variable to policy instruments or other RHS variables are present.
  - If used for more than one period, the ARMA(p,q) must not be re-estimated after imposing the shock, otherwise impacts will be affected by parameter changes.

### First example: Dominica — limited data, no feedback effects
- Context:
  - Small open economy where external factors (USA real GDP, real WTI crude oil price) are treated as exogenous to Dominica.
  - Baseline annual nowcast for Dominica’s real GDP uses four higher-frequency indicators: real domestic imports (dm_im, monthly), tourism expenditures (dm_vis_e, monthly), primary balance (dm_pb, monthly), return on average equity (dm_roe, quarterly); plus two exogenous variables rgdp_usa (quarterly) and r_wti (monthly).
- Baseline regression (annual, 2001-2021): Dependent variable DLOG(RGDP)
  - Sample: 2001 2021; Included observations: 21
  - Coefficients (Variable — Coefficient — Std. Error — t-Statistic — Prob.):
    - C — -0.023209 — 0.010849 — -2.139346 — 0.0505
    - DLOG(DM_IM) — 0.181250 — 0.046806 — 3.872325 — 0.0017
    - DLOG(DM_VIS_E) — 0.051998 — 0.017337 — 2.999274 — 0.0096
    - D(DM_PB) — 0.000422 — 0.000249 — 1.694686 — 0.1123
    - DM_ROE — 0.000749 — 0.000310 — 2.419824 — 0.0297
    - D(R_WTI) — -0.014054 — 0.009818 — -1.431469 — 0.1742
    - D(RGDP_USA) — 0.009849 — 0.003708 — 2.656161 — 0.0188
  - Goodness-of-fit and diagnostics:
    - R-squared 0.881552
    - Adjusted R-squared 0.830788
    - S.E. of regression 0.023466
    - F-statistic 17.36583; Prob(F-statistic) 0.000009
    - Durbin-Watson stat 2.288516
- Example ARMA model for dm_vis_e (quarterly, 2000Q2-2021Q4): preferred ARMA(2,1) with CRISIS dummy, DLOG(RGDP_USA), R_WTI
  - Sample: 2000Q2 2021Q4; Included observations: 87
  - Coefficients (Variable — Coefficient — Std. Error — t-Statistic — Prob.):
    - C — 51.36789 — 10.46747 — 4.907385 — 0.0000
    - CRISIS — -7.346555 — 4.069085 — -1.805456 — 0.0748
    - DLOG(RGDP_USA) — 115.8278 — 37.74850 — 3.068408 — 0.0029
    - R_WTI — -2.209432 — 5.524214 — -0.399954 — 0.6903
    - AR(1) — 1.459534 — 0.487461 — 2.994153 — 0.0037
    - AR(2) — -0.536291 — 0.447560 — -1.198255 — 0.2344
    - MA(1) — -0.412037 — 0.561082 — -0.734362 — 0.4649
    - SIGMASQ — 79.11120 — 9.889257 — 7.999711 — 0.0000
  - Model fit:
    - R-squared 0.833063; Adjusted R-squared 0.818271; S.E. of regression 9.333941
    - Akaike info criterion 7.412926; Schwarz criterion 7.639676; Durbin-Watson stat 1.994653
- Steps to produce nowcast and scenarios:
  - Convert monthly data to quarterly to match rgdp_usa; extend quarterly forecasts to 2022Q4 using ARMA(p,q) models, then convert to annual frequency by averaging quarterly observations.
  - Use extended annual high-frequency series in baseline regression estimated with 2001-2021 to nowcast 2022 RGDP.
  - Create alternative scenarios for exogenous external factors (e.g., 10 percent increase in real oil prices for 2022; 1 percent increase in USA real GDP for 2022), recalculate high-frequency forecasts where affected, recompute nowcast, and compare to baseline.
- Dominica scenario results (one-period ahead impacts):
  - Baseline real GDP nowcast for 2022: 9.55 percent year-on-year growth (or 1217.62 million ECU dollars).
  - Given a 10 percent increase in real oil prices for 2022:
    - Revised nowcast: 9.43 percent (or 1216.35 million ECU dollars).
    - Interpretation: one-period ahead impact of higher oil prices is rather small.
  - Given a 1 percent increase in USA real GDP during 2022:
    - Revised nowcast: 11.34 percent year-on-year growth (or 1237.51 ECU dollars) compared to baseline 9.55 percent.
- Multi-period extension:
  - Because USA real GDP and global oil prices are exogenous to Dominica, the baseline regression can provide multi-period estimates; high-frequency variables must be extended (e.g., to 2023) while keeping baseline parameter estimates unchanged.

### Second approach: non-parametric structural estimator (Ouliaris and Pagan, 2022)
- Method overview:
  - A non-parametric structural estimator estimates responsiveness of a nowcast to a structural shock without requiring full identifying restrictions.
  - Procedure: simulate structural parameters for the structural equation of interest to estimate the structural shock, then regress other variables on the estimated shock from a reduced-form VAR to obtain responses.
  - Advantages in nowcasting context:
    - Low number of parameters relative to VAR.
    - Appropriately agnostic about identifying restrictions underlying true impulse responses.
- Triangular SVAR illustration (three-variable system):
  - Structural equation for high-frequency variable x1t:
    - x1t = α12 x2t + α13 y3t + ... + η1t
  - Reduced-form equations:
    - x2t = ... + ρ1 η1t + v2t
    - y3t = ... + ρ2 η1t + v3t
  - Estimation:
    - Simulate α12 and α13 consistent with economic priors; obtain residuals η̂1t.
    - Regress x2t and y3t on lags of endogeneous variables and η̂1t to estimate ρ1 and ρ2 by OLS; ρ2 is the response of y3t to a unit shock in x1t.
  - If x1t is exogenous, α12 and α13 = 0, and η1t can be replaced by x1t.

### Second example: Tonga — limited data, general feedback effects
- Context:
  - Tonga: small Polynesian economy with frequent natural shocks and heavy reliance on remittances; main partners Australia, New Zealand, USA.
  - Annual real GDP released with 1-2 year lag; semi-annual nowcasting model estimated.
- Baseline semi-annual regression for Tonga (2011S2-2021S1): Dependent variable DLOG(GDPR_CH)
  - Sample (adjusted): 2011S2 2021S1; Included observations: 20
  - Coefficients (Variable — Coefficient — Std. Error — t-Statistic — Prob.):
    - C — -0.000666 — 0.006342 — -0.105059 — 0.9186
    - DLOG(GDPR_CH(-1)) — 0.447579 — 0.221554 — 2.020177 — 0.0741
    - DLOG(TOTAL_AGR) — 0.002912 — 0.006289 — 0.462935 — 0.6544
    - DLOG(CONST_PERR(-1)) — 0.015971 — 0.007837 — 2.037828 — 0.0720
    - DLOG(CONST_PERR(-2)) — 0.011338 — 0.008927 — 1.270036 — 0.2359
    - DLOG(REM_R) — 0.009365 — 0.017558 — 0.533381 — 0.6067
    - DLOG(REM_R(-1)) — 0.004907 — 0.015061 — 0.325794 — 0.7520
    - CRD_GRW(-1) — 0.000128 — 0.000295 — 0.433359 — 0.6750
    - DLOG(TRV_RECR) — 0.005767 — 0.017392 — 0.331609 — 0.7478
    - DLOG(AUS_GDP(-1)) — 0.191812 — 0.424280 — 0.452088 — 0.6619
    - CRISIS — -0.003879 — 0.010082 — -0.384716 — 0.7094
  - Fit and diagnostics:
    - R-squared 0.728796; Adjusted R-squared 0.427458; S.E. of regression 0.009060
    - Akaike info criterion -6.268467; Prob(F-statistic) 0.099773; Durbin-Watson stat 1.325430
- Problem:
  - Large number of explanatory variables with limited observations (20 semi-annual observations) makes conventional structural approaches infeasible.
- Application of non-parametric approach to assess impact of tourism shock on nowcast:
  - Structural equation for travel receipts (trv_recr) specified as function of current real GDP (gdpr_ch), remittances (rem_r), total agricultural production (total_agr) plus lagged terms and constant; structural error η1t captures the shock to travel receipts.
  - Procedure:
    - Simulate β1, β2, β3 from a random distribution (e.g., uniform), estimate residuals from the structural equation to obtain η̂1t for each draw.
    - Use reduced-form VAR equation for real GDP with η̂1t as regressor to estimate ρ2 by OLS; ρ2 is the response of the nowcast of real GDP to a positive unit shock in travel receipts.
    - Search over feasible space for β1, β2, β3 and compute summary statistics for ρ2 (mean, median, max, min) to obtain robust estimates for adjusting baseline nowcast given assumed shock size.
  - Implementation choices in the Tonga example:
    - Regression in equation (4) uses data regressed on a constant and crisis dummy.
    - To avoid unintended constraints, assumed range (−∞, ∞) for β1, β2, β3.
    - Random values di for β1, β2, β3 were drawn from uniform (−1, 1) and transformed to (−∞, ∞) using 1/(1−lal(di)). One million estimates of η1t and ρ2 were generated.
- Empirical result reported in source:
  - The estimates for ρ2 from equation (5) suggest that, on average, there is a positive relationship between the tourism sector and real GDP.

*Source: IMF Working Paper — Introduction (wpiea2023153-print-pdf)*

### 0.56 percent,  a median of 0.24 percent and an interquartile range of [0, 1.12]. These statistics

### Assessing the Impact of Policy Changes on a Nowcast

### Key statistical sensitivity estimate
- The sensitivity of the baseline nowcast for real GDP to a unit shock arising from the tourism sector (i.e., travel receipts) one period ahead is summarized by: 0.56 percent, a median of 0.24 percent and an interquartile range of [0, 1.12].
- These statistics can be used to estimate the sensitivity of the baseline nowcast for real GDP to a unit shock from the tourism sector one period ahead.

### Methodologies proposed
- Partial equilibrium approach:
  - Based on an AAAAAA(p,p?) model as presented in the paper (notation in source: 퐴퐴퐴퐴퐴퐴퐴퐴(푝푝,푞푞)).
  - Regression-based and produces one period estimates of the impact of a policy change.
- Non-parametric structural estimator:
  - Based on the sign-restricted methodology.
  - Non-parametric and produces one period estimates of the impact of a policy change.
- The two approaches:
  - Can be shown to be theoretically equivalent when estimating the impact of exogenous variables.
  - Both yield one period estimates useful for policy analysis.

### Practical considerations for implementation
- Work with residuals after regressing on a constant and the crisis dummy (residuals of 푑푑푙푙푑푑푙푙(푡푡푡푡푣푣_푡푡푟푟푟푟푡푡/푡푡) as specified).
- All required variables in the regression are included with a single lag.
- For the non-parametric sign restrictions, the endpoints of the range (i.e., −∞ and ∞) are excluded.

### Conclusion and policy relevance
- The sensitivity of the baseline nowcast for a low frequency variable from a structural shock to a high frequency explanatory variable is critical information for policymakers.
- The paper provides two regression-based tools that allow policymakers to estimate the one period impact of policy changes on nowcasts, illustrated with a country-specific example.

*IMF Working Paper No. WP/2023/153 — Assessing the Impact of Policy Changes on a Nowcast*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023153-print-pdf.pdf_