## 7.1 Output (wp17109)

## Source details

**Canonical URL:** [7.1 Output (wp17109)](https://www.imf.org/-/media/files/publications/wp/2017/wp17109.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17109.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17109.pdf.json)

---

### Introduction and research design
- Research question: whether economic programs supported by international financial institutions (IFIs) harm or help the countries involved.
- Two core empirical challenges:
  - Adverse selection and reverse causality: countries apply for IFI-support because they face persistent economic problems.
  - Counterfactual construction: difficulty finding comparable data for similarly severe crises without IFI-support.
- Additional complication: IMF-programs often include direct financial transfers, complicating identification of effects of IMF advice, monitoring, and approval versus financing.

### The Policy Support Instrument (PSI)
- Characteristics:
  - Established in October 2005.
  - Non-financial IMF-instrument available to countries with no current or prospective Balance of Payments (BoP) needs.
  - PSIs can only be granted to countries that are “PRGT-eligible”.
    - “PRGT” = “Poverty Reduction and Growth Trust”.
    - Per capita income cutoff: IDA operational cutoff, located at a per capita Gross National Income level of US$1,215 in 2015.
    - Market access criterion evaluated case-by-case.
  - Purpose: offer IMF approval for policies without a borrowing arrangement; help design programs for macroeconomic stability and debt sustainability; encourage structural reforms; signal IMF endorsement to donors, multilateral development banks, and market participants.
  - Described as replicating a traditional IMF-program without financing (Taylor, 2006).
- PSI usage to date (as reported):
  - Adopted by seven countries (all in sub-Saharan Africa): Cabo Verde, Mozambique, Nigeria, Rwanda, Senegal, Tanzania, and Uganda.
- Advantages for evaluation:
  - Non-financial nature isolates effects of Fund advice, monitoring, and approval.
  - Targeting countries without severe BoP needs mitigates adverse selection.

### Methodology: Synthetic Control Method (SCM)
- SCM overview:
  - Constructs a synthetic control as a weighted-average of untreated units to match pre-intervention characteristics.
  - Treatment effect estimated by comparing post-intervention evolution of treated unit to its synthetic control.
  - Applicable with a single treated unit or multiple (here: seven treated countries; donor pool: 39 countries).
- Formal setup and estimator (preserved notation and expressions):
  - Treated unit i = 1; treatment at time T0.
  - Y^N_it (outcome in absence of intervention); Y^I_it (outcome with intervention).
  - Treatment effect α_1t = Y^I_1t − Y^N_1t, t ≥ T0.
  - Factor model: Y^N_it = δ_t + Z_i θ_t + λ_t μ_i + ε_it.
  - Synthetic control estimator: ˆY^N_1t = Σ_{i=2}^{J+1} w_i Y_it.
  - Weight choice: minimize min_W ||X_1 − X_0 W||_V subject to w_i ≥ 0 and Σ w_i = 1; V chosen to minimize mean-squared prediction error.
  - Treatment effect estimate: ˆα_1t = Y^I_1t − ˆY^N_1t.
- Pre-treatment fit assessment:
  - RMSPE = √(1/T0 Σ_{t=1}^{T0} (Y_1t − Σ_{j=2}^{J+1} w*_j Y_jt)^2).
  - benchmark RMSPE = √(1/T0 Σ_{t=1}^{T0} (Y_1t)^2).
  - Fit Index = RMSPE / benchmark RMSPE.
  - Fit Index of zero = perfect fit; Fit Index > 1 indicates poor fit (authors note Fit Index > 1 never occurs in presented cases).

### Data, covariates, donor pool, and specification
- Primary outcomes:
  - real GDP per capita (growth) — IMF’s World Economic Outlook database; expressed in 2010 US dollars.
  - Consumer Price Index (CPI; 2005=100) — IMF’s World Economic Outlook database; rebased to 2005=100 for all countries.
  - Capital stock (total and foreign-owned) — Penn World Tables; updated Lane and Milesi-Ferretti (2007) dataset.
- Data frequency and sample periods:
  - Annual frequency; generally span 1992 through 2015; Rwanda starts in 1995 (exclude 1994); capital stock series end in 2014.
- Covariates (overview, preserved terminology):
  - Investment rate (gross fixed capital formation as a share of GDP).
  - Economic openness (exports + imports as a share of GDP).
  - Population density.
  - Sectoral share of agriculture.
  - Sectoral share of industry.
  - Secondary school enrolment rate.
  - Tertiary school enrolment rate.
  - Absolute value of latitude.
  - Additional for CPI analyses: 7-way classification of the de facto exchange rate regime; measure of central bank governor turnover (proxy for central bank independence).
  - For capital stock analyses: index of capital account openness (Chinn and Ito (2006)).
- Pre-intervention outcome predictors:
  - Real GDP per capita: values in years 1995, 2000, and 2005.
  - CPI: values in years 1995 and 2000.
  - Capital stock: values in years 1995, 2000, and 2005.
- Donor pool:
  - All countries classified as “developing” in the World Economic Outlook database that did not have any kind of IMF-program in place over 1992-2016 → 39 potential control countries.
- Treated countries and rounded treatment-start years (annual data rounding):
  - Cabo Verde (August 2006-February 2012) → treated from 2007.
  - Mozambique (June 2007-June 2016) → treated from 2007.
  - Nigeria (October 2005-October 2007) → treated from 2006.
  - Rwanda (June 2010-December 2016) → treated from 2010.
  - Senegal (November 2007-June 2018) → treated from 2008.
  - Tanzania (February 2007-July 2017) → treated from 2007.
  - Uganda (February 2006-June 2016, extended to June 2017) → treated from 2006.
- Placebo procedure:
  - Iteratively treat donor-pool countries; include only placebo runs with Fit Index ≤ treated country’s Fit Index (or the five best-fitting placebos if fewer than five meet criterion).

### Main results — Output (Real GDP per capita)
- SCM findings summary:
  - All seven PSI-countries outperform their SCM-constructed counterfactuals post-treatment.
  - Treated countries experienced per capita real GDP growth rates about 1 percentage point higher than their synthetic equivalents in the post-treatment period.
- Table of average annual real GDP per capita growth rates post-treatment (exact values preserved):
  - Cabo Verde: Actual 1.8% | Synthetic Control 1.3% | Treatment effect +0.5%
  - Mozambique: Actual 3.9% | Synthetic Control 2.2% | Treatment effect +1.7%
  - Nigeria: Actual 3.7% | Synthetic Control 0.7% | Treatment effect +3.0%
  - Rwanda: Actual 4.4% | Synthetic Control 3.2% | Treatment effect +1.2%
  - Senegal: Actual 1.0% | Synthetic Control 0.4% | Treatment effect +0.6%
  - Tanzania: Actual 4.0% | Synthetic Control 2.9% | Treatment effect +1.1%
  - Uganda: Actual 3.0% | Synthetic Control 1.8% | Treatment effect +1.2%
- Robustness and notes:
  - Placebo exercises: output in some placebo countries ambiguous or negative, a pattern not observed for actual PSI-countries; PSI treatment effects are sizable relative to placebos.
  - Excluding largest-weight donor country typically does not change results (exception: Cabo Verde — treatment effect disappears when largest-weight donor excluded).
  - Results robust to inclusion of rule-of-law index (donor pool shrinks; pre-treatment Fit Index worsens).
  - Nigeria: SCM sometimes unable to construct reasonable synthetic control and is disregarded in parts of the analysis; when measurable, Nigeria shows largest growth treatment effect (+3.0 percentage points).

### Main results — Price level (CPI inflation)
- CPI analysis summary:
  - Containing inflation is an explicit PSI goal.
  - In all PSI-countries except Uganda, annual inflation rates were lower than in their synthetic control — typically by about 3 percentage points — while growth tended to be higher.
- Table of average annual CPI-inflation rates post-treatment (exact values preserved):
  - Cabo Verde: Actual 2.5% | Synthetic Control 5.8% | Treatment effect -3.3%
  - Mozambique: Actual 5.9% | Synthetic Control 14.2% | Treatment effect -8.3%
  - Nigeria: Actual 10.0% | Synthetic Control 11.5% | Treatment effect -1.5%
  - Rwanda: Actual 4.1% | Synthetic Control 7.4% | Treatment effect -3.3%
  - Senegal: Actual 0.5% | Synthetic Control 2.7% | Treatment effect -2.2%
  - Tanzania: Actual 9.7% | Synthetic Control 15.1% | Treatment effect -5.4%
  - Uganda: Actual 9.1% | Synthetic Control 4.8% | Treatment effect +4.3%
- Uganda-specific observation:
  - Inflation episode spanning 2011-12: inflation equaled 18.7 and 14.0 percent respectively.
  - Note: 2011 PSI review for Uganda not completed; Acting Chair’s Summing Up (February 11, 2011) highlighted fiscal and monetary uncertainty and urged renewed fiscal discipline and limits to central bank financing.
- Robustness:
  - Placebo effects tend to be small relative to treatment effects, except for Senegal.
  - Excluding largest-weight donors yields similar results.

### Main results — Capital stock and foreign-owned capital stock (FDI)
- Total capital stock (percent of GDP):
  - Most treated countries saw substantial increases in capital stocks during PSI periods.
  - Comparison to synthetic controls suggests PSI adoption does not generate a large robust effect on overall capital accumulation.
  - Apparent positive effects for Cabo Verde, Tanzania, and Uganda are sensitive to covariate specification and not robust.
- Foreign-owned capital stock:
  - Evidence of positive treatment effects on foreign-owned capital stock in most treated countries.
  - All treated countries (except Tanzania) accumulated more foreign-owned capital post-PSI than their synthetic controls (Senegal’s effect did not persist past 2014).
  - Results robust to different regression specifications.
- Interpretation:
  - PSIs appear to signal positively to foreign investors, catalyzing FDI.
  - FDI can drive growth via knowledge transfers and improved management practices, even if overall investment is not strongly raised.

### Discussion — interpretation, alternative explanations, and robustness
- Potential biases from prior traditional IMF programs (three cases considered):
  1. Traditional IMF programs had no impact → no bias.
  2. Traditional IMF programs had a positive impact → would bias PSI results downward.
  3. Traditional IMF programs had a negative impact → could make positive post-PSI effects reflect termination of traditional programs rather than PSI adoption.
- Empirical mitigation of bias concerns:
  - Six of seven PSI countries also experienced traditional IMF programs, mainly prior to PSI adoption.
  - Mozambique, Senegal, Tanzania operated disbursing IMF programs in parallel to a PSI; if traditional programs had negative effects, these should show smaller treatment effects — they do not.
  - Nigeria (no disbursing IMF programs over 1992-2015) shows results consistent with other PSI countries and, when measurable, the largest growth treatment effect (+3.0 percentage points), suggesting any bias likely attenuates measured PSI effects.
- Possible reasons for divergence from earlier studies:
  - IMF programs may have become more effective over time (earlier studies used data spanning 1970-2000).
  - Prior studies focused on funded IMF programs often launched in severe crises; PSIs allow evaluation of IMF advice absent direct financing.
  - Financial assistance could introduce negative macroeconomic effects (e.g., real appreciation/Dutch Disease, reduced reform incentives).
- Limitations:
  - Results apply solely to PSIs (non-financial IMF programs) and do not automatically generalize to IMF programs with financial components or crisis contexts.
  - Additional research needed on IMF-advice effectiveness in crises and net effects of financial assistance.

### Appendix B — Comparative PSM analysis of IMF program effectiveness over time
- Method: Propensity Score Matching (PSM) using probit to estimate probability of program entry; treatment effects computed as differences in matched averages.
- Probit variables (lagged):
  - level of real GDP per capita;
  - growth rate of real GDP per capita;
  - current account balance (percent of GDP);
  - overall fiscal balance (percent of GDP);
  - growth rate of terms-of-trade;
  - percentage point change in international reserves (percent of GDP);
  - dummy for banking or currency crisis.
- Probit regression results (selected exact coefficients and statistics preserved):
  - Full sample (Column 1; pseudo-R2 = 0.1030; obs: 3,210):
    - real GDP/cap (level): −0.0000385 ∗∗∗ (−7.85)
    - real GDP/cap (growth): −0.0084595 ∗ (−1.90)
    - current account balance: −0.0092706 ∗∗ (−2.46)
    - fiscal balance: −0.0127646 ∗∗ (−2.14)
    - terms-of-trade (growth): 0.0028846 (1.54)
    - intl. reserves (change): −0.0005187 (−1.04)
    - banking or currency crisis?: 0.5713349 ∗∗∗ (3.91)
  - LIDC sample (Column 2; pseudo-R2 = 0.0349; obs: 948):
    - real GDP/cap (level): −0.0001871 ∗∗∗ (−4.44)
    - fiscal balance: −0.0161166 ∗ (−1.88)
    - banking or currency crisis?: 0.4276361 ∗∗ (2.13)
- PSM average treatment effect on treated (LIDCs; 5-year averages; exact values preserved):
  - 1980-2015 (Column 1; obs: 732):
    - real GDP/cap (growth): −1.165666 ∗∗ (−2.11)
    - inflation: 0.6269521 (0..81)
  - 2000-2015 (Column 2; obs: 732):
    - real GDP/cap (growth): −0.308217 (−0.45)
    - inflation: 0.2236667 (0.33)
- Interpretation:
  - 1980-2015: IMF programs associated with statistically significant decrease in real GDP per capita growth (−1.165666 ∗∗).
  - 2000-2015: negative growth effect disappears (−0.308217, not significant); inflation effects remain insignificant and closer to zero.
  - Suggested explanation: improvements after the 1999 PRGT-reforms emphasizing poverty reduction and country ownership.

### Synthetic-control fit indices and covariate balance (selected exact Fit Index values and covariate comparisons)
- Fit Index (real GDP per capita):
  - Cabo Verde: 0.020
  - Mozambique: 0.090
  - Nigeria: 0.015
  - Rwanda: 0.075
  - Senegal: 0.028
  - Tanzania: 0.015
  - Uganda: 0.030
- Fit Index (CPI):
  - Cabo Verde: 0.041
  - Mozambique: 0.104
  - Nigeria: 0.108
  - Rwanda: 0.064
  - Senegal: 0.067
  - Tanzania: 0.050
  - Uganda: 0.035
- Fit Index (capital stock):
  - Cabo Verde: 0.016
  - Mozambique: 0.065
  - Nigeria: 0.039
  - Rwanda: 0.075
  - Senegal: 0.023
  - Tanzania: 0.019
  - Uganda: 0.021
- Fit Index (foreign-owned capital stock):
  - Cabo Verde: 0.181
  - Mozambique: 0.149
  - Nigeria: 0.291
  - Rwanda: 0.153
  - Senegal: 0.078
  - Tanzania: 0.206
  - Uganda: 0.091
  - Note: "Nigeria has to be disregarded in this analysis since no “fitting” synthetic control can be constructed from the donor pool (see Figure 8)."
- Selected covariate balances (treated / synthetic; exact values preserved):
  - Real GDP per capita covariates — investment rate:
    - Uganda: 33.0054 / 29.07862
    - Cabo Verde: 19.27152 / 19.54908
    - Mozambique: 18.09357 / 23.58386
    - Nigeria: 19.59305 / 18.79299
    - Rwanda: 21.99165 / 22.22125
    - Senegal: 22.19416 / 22.48265
    - Tanzania: 24.06709 / 24.09635
  - CPI covariates — exchange rate regime:
    - Uganda: 2.4 / 2.3922
    - Cabo Verde: 6.066667 / 4.744067
    - Mozambique: 6 / 4.293182
    - Nigeria: 5.133333 / 5.131267
    - Rwanda: 1 / 1.318
    - Senegal: 6.866667 / 4.5462
    - Tanzania: 7 / 5.766
  - Capital stock covariates — capital stock in pct of GDP (2005):
    - Uganda: 687.4423 / 695.3142
    - Cabo Verde: 339.3521 / 373.453
    - Mozambique: 437.9172 / 467.3235
    - Nigeria: 330.6231 / 348.0999
    - Rwanda: 883.7039 / 884.2657
    - Senegal: 1438.64 / 1463.466
    - Tanzania: 635.4164 / 638.6998
  - Foreign-owned capital stock covariates — for. cap. stock in pct of gdp (2005):
    - Uganda: 33.04366 / 34.19779
    - Cabo Verde: 35.00573 / 33.55399
    - Mozambique: 2.990634 / 3.342479
    - Nigeria: 15.34797 / 15.56053
    - Rwanda: 26.21858 / 27.03011
    - Senegal: 18.39113 / 17.99228

### Conclusion (exact phrasing preserved where applicable)
- Primary SCM finding: "Applying the SCM to PSI-treated countries suggests that this type of IMF program has had a positive effect on economic development in those countries."
- Main quantified findings:
  - Adopting a PSI is associated with increased economic growth: treated countries tend to add one percentage point in average annual GDP per capita growth in the years following adoption.
  - Adopting a PSI is associated with lower inflation: inflation decreases by about 3 percentage points per year in treated countries (except Uganda).
  - PSIs are associated with increased foreign investment (FDI), which can bring knowledge transfers and improved management practices.
- Caveat:
  - Findings are specific to PSIs and do not necessarily extrapolate to IMF programs that include financial assistance or are implemented during severe crises.
- Overall implication:
  - For the case of PSIs, adopting countries might experience substantial macroeconomic benefits.

*Source: wp17109 - 7.1 Output (IMF Working Paper content provided).*

### 7.1    Output    .......................................................................................................

### wp17109 - 7.1    Output    .......................................................................................................

### Introduction
- Research question: whether economic programs supported by international financial institutions (IFIs) harm or help the countries involved.
- Two core empirical challenges highlighted:
  - Adverse selection and reverse causality: countries apply for IFI-support because they face persistent economic problems, making causality difficult to establish.
  - Counterfactual construction: every economic crisis has unique elements and it is hard to find comparable data from countries experiencing similarly severe crises without IFI-support.
- Additional complication: IMF-programs often include direct financial transfers, making it hard to disentangle the effects of IMF advice, monitoring, and approval from the effects of financing.

### The Policy Support Instrument (PSI)
- PSI characteristics:
  - Established in October 2005.
  - A non-financial IMF-instrument available to countries with no current or prospective Balance of Payments (BoP) needs (a necessary condition for all other IMF programs).
  - PSIs can only be granted to countries that are “PRGT-eligible”.
    - “PRGT” refers to “Poverty Reduction and Growth Trust”.
    - The per capita income cutoff is set at the IDA operational cutoff, which was located at a per capita Gross National Income level of US$1,215 in 2015.
    - Evaluation of the market access criterion is done on a more case-by-case basis.
  - Purpose: offer countries the Fund’s approval for their economic policies without entering a borrowing arrangement; help design programs to ensure macroeconomic stability and debt sustainability; encourage structural reforms to remove constraints on growth and poverty reduction; signal IMF endorsement to donors, multilateral development banks, and market participants.
  - Described as replicating a traditional IMF-program without financing (Taylor, 2006).
- PSI usage to date (as reported):
  - Adopted by seven countries (all in sub-Saharan Africa): Cabo Verde, Mozambique, Nigeria, Rwanda, Senegal, Tanzania, and Uganda.
- Advantages for evaluation:
  - Non-financial nature allows analysis of effects of Fund advice, monitoring, and approval separate from direct financial assistance.
  - Targeting countries without severe BoP needs mitigates adverse selection, facilitating counterfactual construction.

### Methodology: Synthetic Control Method (SCM)
- SCM summary:
  - First employed in Abadie and Gardeazabal (2003) and extended by Abadie, Diamond, and Hainmueller (2010).
  - Constructs a synthetic control as a weighted-average of units that did not undergo treatment over the sample period.
  - Weights chosen so relevant economic characteristics of the synthetic control match the treated unit in the pre-intervention period.
  - Treatment effect estimated by comparing post-intervention evolution of the treated unit to its synthetic control.
- Application in this study:
  - “Treatment” = adoption of a PSI.
  - Synthetic controls constructed from a group of “untreated” developing countries that did not have any kind of IMF-program in place over the sample period.
  - Relation to other methods: related to Rubin-style matching methods; SCM can be applied even with a single treated country. In this study there are seven treated countries and a donor pool containing 39 countries.

### Main results and substantive findings
- Medium-run effects of PSI adoption (as reported):
  - Output/growth: following PSI adoption, countries tend to grow faster than their synthetic controls by about 1 percentage point per year.
  - Inflation: treated countries are characterized by lower rates of inflation by about 3 percentage points per year.
  - Capital stock: although capital stocks increased significantly in most countries during the PSI period, the results suggest this development is not caused by PSI-treatment.
  - Foreign investment: evidence indicates PSIs have stimulated foreign investment.
    - Potential channels: foreign investment may bring knowledge transfers and better management practices that could partly explain positive effects on output.

### Study itinerary (organization)
- Section 2: related literature.
- Section 3: documentation of types of IMF-engagement for the PSI-countries.
- Section 4: core contents of the PSIs launched to date.

*Source: wp17109 - 7.1    Output    .......................................................................................................*

### Section 5 goes on to describe the Synthetic Control Method, which we use to construct

### wp17109 - Section 5 goes on to describe the Synthetic Control Method, which we use to construct

### Related literature
- Comprehensive overviews: Dreher (2006, 2009) and Steinwand and Stone (2008).
- Early cross-country program comparisons: Reichmann and Stillson (1978); Donovan (1981, 1982). Early studies typically conclude IMF-programs successful in stabilizing the economy but are criticized for not controlling for reverse causality and weak counterfactual construction.
- Regression-based and selection-correcting approaches:
  - Dicks-Mireaux, Mecagni, and Schadler (2000): use the General Evaluation Estimator (GEE) to construct counterfactuals via policy reaction functions from non-program countries; find positive effects on growth and the debt-service ratio, no significant effect on inflation; diagnostic tests question reliability.
  - Przeworski and Vreeland (2000) and Atoyan and Conway (2006): use a dynamic Heckman selection model; find IMF-program participation lowers growth while the program is in place, with growth picking up post-program. Atoyan and Conway (2006) confirm with propensity-score matching.
  - Barro and Lee (2005): exploit variation in IMF loan exposure (quota, professional IMF-staff, connectedness) with instrumental variables; find higher IMF loan-participation reduces economic growth, no significant effects on investment, inflation, government consumption, and openness.
  - Dreher (2006): similar IV results; negative growth effects mitigated by compliance with conditionality.
  - Bas and Stone (2014): document adverse selection into IMF programs; taking adverse selection into account yields more favorable impact on growth but large heterogeneity—governments most eager to participate often experience least benefit.
  - Binder and Bluhm (2017): taking adverse selection into account, IMF programs only boost output if accompanied by institutional improvements.
- Studies on inequality, poverty, development: Garuda (2000); Easterly (2003); Hajro and Joyce (2009); mixed results. For PRGT-funded programs: IMF (2012) finds parity with non-seekers; Oberdabernig (2013) and Lang (2016) find no increase or decreases in inequality. CGD (2007) and Clements, Gupta, and Nozaki (2013) report increases in social spending under PRGT-funded programs.
- Contribution of this paper: introduces the Synthetic Control Method (SCM) focused on non-disbursing PSIs to mitigate reverse causality and to isolate effects of Fund advice, monitoring, and approval without financial disbursements.

### PSI-countries and the IMF
- Seven countries adopted PSIs (all in sub-Saharan Africa): Cabo Verde, Mozambique, Nigeria, Rwanda, Senegal, Tanzania, and Uganda.
- Sample period: full sample period 1992-2015 (Rwanda starts in 1995 to exclude the 1994-genocidal episode).
- Observation: all countries apart from Nigeria experienced a BoP-need at some point over 1992-2015 and entered funded IMF programs; in none of these cases was the BoP-need accompanied by an acute economic crisis.
- PSIs typically consist of multiple multi-year programs launched in succession.
- For Mozambique, Senegal, and Tanzania, PSIs co-existed for part of the period with a disbursing IMF program (temporary BoP-need). These disbursing programs were relatively short-lived, with BoP-needs caused by external factors.
- Note on potential bias: presence of earlier programs may bias results, but authors argue in Section 8 the bias likely runs opposite to the results found.
- Distinction: disbursing programs primarily address BoP-problems; PSI main objective is to enhance growth and reduce poverty.
- Focus of analysis: impact of PSIs on growth and inflation outcomes (inflation tax mainly carried by the poor; Easterly and Fischer (2001)).

### Core contents of PSI-programs
- Policy goal (IMF, 2016: 113): PSI-supported programs aim to maintain or consolidate:
  - (i) strong and durable poverty reduction and growth,
  - (ii) low or moderate inflation,
  - (iii) sustainable fiscal and current account balances,
  - (iv) limited debt vulnerabilities,
  - (v) adequate international reserves,
  - (vi) sufficient policy and institutional capacity to implement appropriate macroeconomic policies.
- Conditionality structure: combination of quantitative assessment criteria and structural measures within member country control.
- Quantitative conditionality examples:
  - Floors on international reserves of the central bank,
  - Ceilings on monetary targets,
  - Limits on domestic financing of the government,
  - Constraints on external debt accumulation,
  - Floors on social spending,
  - Ceilings on accumulation of new domestic arrears,
  - Floors on government revenue.
- Structural conditionality focus areas and country examples:
  - Revenue measures: streamlining tax exemptions and incentives (Cabo Verde); implementation of customs tariff regimes (Nigeria); limitation of tax cash payments (Senegal); Rwanda: improvements to tax form submissions (2010), decrease VAT exemptions, revision of property taxation, new tax regimes for agriculture and mining (2013).
  - Expenditure management: restrictions on contracting new non-concessional financing (Mozambique); adoption of medium-term expenditure framework (Cabo Verde); guidelines to assess rate-of-return on projects (Senegal); clearance of domestic arrears (Uganda 2006 PSI).
  - Budget and payments transparency (more than one fourth of measures across PSIs): systems and direct bank transfer for salary payments (Mozambique, Uganda); limiting number of accounts around treasury single account (Mozambique, Senegal); financial management information systems in ministries (Rwanda); improved monitoring of social spending (Tanzania); conditionality on publishing information, ex-post audits, cost-benefit evaluations (Senegal).
  - Financial sector reforms (Cabo Verde): expand/formalize Financial Stability Committee, banking law unifying regulatory framework.
  - Legal/regulatory mining and petroleum reforms (Mozambique): adoption of new model contracts.
  - Public enterprise reforms (Nigeria): unbundling national electricity company, opening bids on state telecommunications strategy.
  - Central Bank reform (Rwanda, Tanzania, Uganda): publish quarterly inflation reports and underlying economic assessments (Rwanda); consolidated supervision of commercial banks and strengthen interbank market (Tanzania); actions to ensure capital adequacy (Uganda).
  - Transparency and growth strategy monitoring (Senegal): institutional framework for implementing and monitoring accelerated growth strategy.
- Analytical focus: use SCM to analyze impact of all PSI-programs on key macroeconomic variables.

### The Synthetic Control Method (SCM)
- Origin: developed by Abadie and Gardeazabal (2003) and extended in Abadie, Diamond, and Hainmueller (2010).
- Setup:
  - Time-series data about J+1 units (countries), indexed by i = 1,2,...,J+1.
  - Treated unit: i = 1 undergoes treatment at time T0. All J other units remain untreated and form the donor pool.
  - Outcome variable Y. Let Y^N_it denote Y in unit i at time t in absence of intervention; Y^I_it denote Y when intervention takes place.
  - Treatment effect for treated unit (i = 1): α_1t = Y^I_1t − Y^N_1t, t ≥ T0.
- Factor model for unobserved counterfactual Y^N_it:
  - Y^N_it = δ_t + Z_i θ_t + λ_t μ_i + ε_it.
  - δ_t: common factor; Z_i: vector of observed covariates (not affected by policy intervention); θ_t: associated unknown parameters; λ_t: vector of unobserved factors; μ_i: factor loadings; ε_it: zero-mean error term.
- Synthetic control construction:
  - Construct counterfactual estimate as ˆY^N_1t = Σ_{i=2}^{J+1} w_i Y_it, t ≥ T0.
  - Choose weights w_i to match pre-intervention characteristics: define X_1 (vector of average values of pre-intervention variables for treated unit) and X_0 (same for donor pool).
  - Solve for W* by minimizing min_W ||X_1 − X_0 W||_V = √(X_1 − X_0 W)' V (X_1 − X_0 W)
    subject to w*_i ≥ 0 for i = 2,...,J+1 and Σ_{i=2}^{J+1} w*_i = 1.
  - V is symmetric and positive semi-definite and chosen to minimize mean-squared prediction error for the outcome variable in the pre-treatment period.
  - Treatment effect estimate for t ≥ T0: ˆα_1t = Y^I_1t − ˆY^N_1t.
- Pre-treatment fit assessment:
  - Use the pre-treatment "Fit Index" of Adhikari and Alm (2016) and Adhikari, Duval, Hu, and Loungani (2016).
  - RMSPE = √(1/T0 Σ_{t=1}^{T0} (Y_1t − Σ_{j=2}^{J+1} w*_j Y_jt)^2).
  - benchmark RMSPE = √(1/T0 Σ_{t=1}^{T0} (Y_1t)^2).
  - Fit Index = RMSPE / benchmark RMSPE.
  - Interpretation: Fit Index of zero implies a perfect fit; Fit Index greater than one indicates particularly poor fit (Adhikari, Duval, Hu, and Loungani, 2016); the authors note Fit Index > 1 never occurs in their presented cases.

### Data and regression specifications
- Primary outcomes: real GDP per capita (growth) and Consumer Price Index (inflation). Also analyze capital stock (total and foreign-owned) to explore channels.
- Data frequency and sample periods:
  - Annual frequency.
  - Generally span years 1992 through 2015; Rwanda starts in 1995 (exclude 1994).
  - Capital stock series end in 2014; capital stock analyses span 1992-2014.
- Real GDP per capita:
  - Source: IMF’s World Economic Outlook database.
  - Expressed in 2010 US dollars.
- Covariates included (inspired by Abadie and Gardeazabal (2003) and Barro and Sala-i-Martin (1995); overview in Table 1 of source):
  - Investment rate (gross fixed capital formation as a share of GDP) — source: World Economic Outlook database.
  - Economic openness (exports + imports as a share of GDP) — source: World Development Indicators.
  - Population density — source: World Development Indicators.
  - Sectoral share of agriculture — source: World Development Indicators.
  - Sectoral share of industry — source: World Development Indicators.
  - Secondary school enrolment rate — source: World Development Indicators.
  - Tertiary school enrolment rate — source: World Development Indicators.
  - Absolute value of latitude (added to capture institutional effects; following Cavallo et al. (2013)).
- Pre-intervention outcome predictors (following Kaul et al. (2016) recommendation to limit number):
  - For real GDP per capita: real GDP per capita in years 1995, 2000, and 2005.
  - For CPI: CPI in years 1995 and 2000.
- CPI data:
  - Source: IMF’s World Economic Outlook database.
  - Rebasing: all countries have CPI = 100 in year 2005.
  - Additional covariates for CPI analyses: 7-way classification of the de facto exchange rate regime (inspired by Ghosh, Ostry, and Tsangarides (2010)); measure of central bank governor turnover as proxy for central bank independence.
- Capital stock measures:
  - Total capital stock: Penn World Tables.
  - Foreign-owned capital stock: updated and extended Lane and Milesi-Ferretti (2007) dataset.
- Data limitations noted:
  - Unable to analyze fiscal variables or direct measures of inequality (e.g., Gini coefficients) due to data constraints.

*Source: wp17109 - Section 5 goes on to describe the Synthetic Control Method, which we use to construct (IMF Working Paper PDF content provided).*

### 2014.  We follow Sanso-Navarro (2011) – who in turn based his choice on Blonigen et al.

### wp17109 - 2014.  We follow Sanso-Navarro (2011) – who in turn based his choice on Blonigen et al.

### Covariates and predictors
- Covariates included (following Sanso-Navarro (2011) and Blonigen et al. (2007)):
  - population size (to reflect potential market size)
  - openness (the sum of exports and imports as a share of GDP)
  - secondary school enrolment rate
  - tertiary school enrolment rate
  - index of capital account openness (see Chinn and Ito (2006))
  - absolute value of latitude
  - note: the latter two series proxy for “host-country investment costs”
- Pre-treatment outcome-variable predictors:
  - capital stock in 1995, 2000, and 2005
- Additional covariate notes:
  - CPI equals 100 for every country in the year 2005, so that observation is not used as a predictor.
  - Blonigen et al. (2007) use a composite risk index, but that variable is not available for most countries in the present study.
- Table 1 covariate overview (as used across outcome variables):
  - Real GDP per capita: investment rate, population size, central bank independence, population density, share of agriculture, share of industry, sec school enr rate, tert school enr rate, latitude (absolute value)
  - CPI (2005=100): exchange rate regime, openness, sec school enr rate, tert school enr rate, latitude (absolute value)
  - Capital stock (total and foreign): investment rate, openness, capital account openness, sec school enr rate, tert school enr rate, latitude (absolute value)

### Donor pool and SCM construction
- Donor pool definition:
  - all countries classified as “developing” by the World Economic Outlook database that did not have any kind of IMF-program in place over the period 1992-2016
  - yields 39 potential control countries
- Synthetic Control Method (SCM) details:
  - synthetic control constructed for each treated country from a weighted average of donor-pool countries to match covariates and pre-treatment outcomes
  - algorithm minimizes mean-squared prediction error for the outcome variable in the pre-treatment period, prioritizing covariates important for minimizing prediction error
  - placebo exercises: iteratively treat donor-pool countries; only placebo runs with Fit Index lower than or equal to the treated country’s Fit Index are included (or the five best-fitting placebos if fewer than five meet the criterion)
- Notes on treated countries and date rounding:
  - Seven PSI adopters (all sub-Saharan Africa): Cabo Verde (August 2006-February 2012), Mozambique (June 2007-June 2016), Nigeria (October 2005-October 2007), Rwanda (June 2010-December 2016), Senegal (November 2007-June 2018), Tanzania (February 2007-July 2017), Uganda (February 2006-June 2016, now extended to June 2017)
  - With annual data, dates are rounded to nearest integer for treatment start (e.g., Cabo Verde treated from 2007, Nigeria from 2006)
  - Cross-country spillovers from sub-Saharan African treated countries to controls are judged minimal

### Results — Output (Real GDP per capita)
- SCM findings summary:
  - All seven PSI-countries outperform their SCM-constructed counterfactuals in the post-treatment period.
  - Treated countries experienced per capita real GDP growth rates about 1 percentage point higher than their synthetic equivalents in the post-treatment period.
- Table 2: Average annual real GDP per capita growth rates post-treatment
  - Cabo Verde: Actual 1.8% | Synthetic Control 1.3% | Treatment effect +0.5%
  - Mozambique: Actual 3.9% | Synthetic Control 2.2% | Treatment effect +1.7%
  - Nigeria: Actual 3.7% | Synthetic Control 0.7% | Treatment effect +3.0%
  - Rwanda: Actual 4.4% | Synthetic Control 3.2% | Treatment effect +1.2%
  - Senegal: Actual 1.0% | Synthetic Control 0.4% | Treatment effect +0.6%
  - Tanzania: Actual 4.0% | Synthetic Control 2.9% | Treatment effect +1.1%
  - Uganda: Actual 3.0% | Synthetic Control 1.8% | Treatment effect +1.2%
- Robustness and placebo notes:
  - Placebo exercises: output in some placebo countries ambiguous or negative, a pattern not observed for actual PSI-countries; PSI treatment effects are sizable relative to placebo effects.
  - Excluding the largest-weight donor country typically does not change results (exception: Cabo Verde, where the treatment effect disappears when the largest-weight donor is excluded).
  - Results robust to inclusion of rule-of-law index as a covariate, though donor pool shrinks and pre-treatment Fit Index worsens.
  - Nigeria: SCM unable to construct a reasonable synthetic control for some analyses and is disregarded in parts of the remaining analysis (but when measurable, Nigeria shows the largest growth treatment effect: +3.0 percentage points).

### Results — Price level (CPI inflation)
- CPI analysis summary:
  - Containing inflation is an explicit PSI goal; analysis focuses on CPI evolution post-treatment.
  - In all PSI-countries except Uganda, annual inflation rates were lower than in their synthetic control — typically by about 3 percentage points — even while growth tended to be higher.
- Table 3: Average annual CPI-inflation rates post-treatment
  - Cabo Verde: Actual 2.5% | Synthetic Control 5.8% | Treatment effect -3.3%
  - Mozambique: Actual 5.9% | Synthetic Control 14.2% | Treatment effect -8.3%
  - Nigeria: Actual 10.0% | Synthetic Control 11.5% | Treatment effect -1.5%
  - Rwanda: Actual 4.1% | Synthetic Control 7.4% | Treatment effect -3.3%
  - Senegal: Actual 0.5% | Synthetic Control 2.7% | Treatment effect -2.2%
  - Tanzania: Actual 9.7% | Synthetic Control 15.1% | Treatment effect -5.4%
  - Uganda: Actual 9.1% | Synthetic Control 4.8% | Treatment effect +4.3%
- Uganda-specific observations:
  - Inflation in Uganda was pushed up by an inflationary episode spanning 2011-12 when inflation equaled 18.7 and 14.0 percent respectively.
  - In 2011, a PSI review for Uganda was not completed; Acting Chair’s Summing Up (February 11, 2011) noted concerns about a supplementary budget, fiscal and monetary uncertainty, and urged renewed fiscal discipline and limits to central bank financing of the deficit.
- Robustness and placebo notes:
  - Placebo effects tend to be small relative to treatment effects, except for Senegal.
  - Excluding largest-weight donors yields similar results, supporting that PSIs helped control inflation.

### Results — Capital stock and FDI
- Total capital stock (percent of GDP):
  - Most treated countries saw substantial increases in capital stocks over the course of their PSI.
  - Comparison to synthetic controls suggests adoption of a PSI does not generate a large robust effect on overall capital accumulation.
  - Apparent positive effects for Cabo Verde, Tanzania, and Uganda are sensitive to covariate specification and not robust.
- Foreign-owned capital stock (FDI):
  - More evidence for a positive treatment effect on foreign-owned capital stock.
  - All treated countries (except Tanzania) accumulated more foreign-owned capital in years following PSI adoption than their synthetic controls (although Senegal’s effect did not persist past 2014).
  - Figures show treatment effects in Cabo Verde, Mozambique, Rwanda, and Senegal (until 2014) large relative to placebo effects.
  - Results robust to different regression specifications.
- Interpretation:
  - PSIs appear to emit a positive signal to foreign investors, catalyzing FDI.
  - While no strong evidence that this raises overall investment (possible crowding out of domestic investment), attracting FDI can drive growth via knowledge transfers and improved management practices.

### Discussion — interpretation, alternative explanations, and robustness
- Potential biases from prior traditional IMF programs (three cases considered):
  1. Traditional IMF programs had no impact on macroeconomic situation → no bias.
  2. Traditional IMF programs had a positive impact → would bias results downward (harder to find additional PSI effects).
  3. Traditional IMF programs had a negative impact → could make positive post-PSI effects reflect termination of traditional programs rather than PSI adoption.
- Empirical evidence mitigating bias concerns:
  - Six of seven PSI countries also went through traditional IMF programs, mainly prior to PSI adoption.
  - Three countries (Mozambique, Senegal, Tanzania) operated disbursing IMF programs in parallel to a PSI; if traditional programs had negative effects, these should show smaller treatment effects — they do not.
  - Nigeria (the “purest” PSI-country, without disbursing IMF programs over 1992-2015) shows results in line with other PSI countries; in measured cases Nigeria exhibits largest growth treatment effect (+3.0 percentage points), suggesting any bias from prior programs likely attenuates measured PSI effects rather than causing spurious positives.
- Reasons results may differ from earlier studies:
  - IMF programs may have become more effective over time; earlier studies used data spanning 1970-2000 whereas IMF operations have evolved (e.g., toward supporting country-led development agendas).
  - Earlier studies focused on IMF programs with financial assistance, often launched in severe crises, making reverse causality a greater concern; PSIs enable analysis of IMF advice absent direct financial assistance.
  - Traditional IMF programs may be less effective than PSIs, or the presence of financial assistance could introduce negative macroeconomic effects (e.g., real appreciation/Dutch Disease, reduced reform incentives).
- Limitations noted:
  - Results apply solely to PSIs (non-financial IMF programs); they do not automatically generalize to IMF programs with financial components or to programs launched during severe crises.
  - Additional research needed to assess IMF-advice effectiveness in crisis contexts and net effects of financial assistance.

### Conclusion
- This paper is the first application of the Synthetic Control Method to evaluate programs operated by an international financial institution (the IMF) using non-financial Policy Support Instruments (PSIs).
- Main findings:
  - Adopting a PSI is associated with increased economic growth: treated countries tend to add one percentage point in average annual GDP per capita growth in the years following adoption.
  - Adopting a PSI is associated with lower inflation: inflation decreases by about 3 percentage points per year in treated countries (except Uganda).
  - PSIs are associated with increased foreign investment (FDI), which can bring knowledge transfers and improved management practices.
- Caveat:
  - Findings are specific to PSIs and do not necessarily extrapolate to IMF programs that include financial assistance or are implemented during severe crises.
- Overall implication:
  - For the case of PSIs, adopting countries might experience substantial macroeconomic benefits.

### Appendix A: donor pool countries (as listed)
- Bahamas*  Equatorial Guinea*  Marshall Islands  Qatar*  Timor-Leste
- Bahrain*  Eritrea  Mauritius*  Samoa  Tonga
- Barbados*  Fiji*  Micronesia  Saudi Arabia*  Trinidad & Tobago*
- Belize*  Iran*  Montenegro  Saint Lucia*  Turkmenistan
- Bhutan*  Kiribati  Myanmar  Sudan*  Tuvalu
- Botswana  Kuwait  Namibia*  Suriname*  Utd Arab Emirates*
- Brunei  Lebanon*  Oman*  Swaziland*  Vanuatu
- Dem. Rep. Congo  Libya  Palau  Syria
- Note: table contains all countries classified as “developing” in the IMF’s World Economic Outlook database while not having any kind of IMF-program in place over the period 1992-2016.

*Source: World Economic Outlook database and own calculations.*

### 2016.  Due to data limitations, not all countries can be included.  Only the countries in bold are considered

### wp17109 - 2016

### Synthetic control results and fit indices (real GDP per capita, CPI, capital stock, foreign-owned capital stock)
- Table A2 (Synthetic controls and fit indices for real GDP per capita)
  - Fit Index by treated country:
    - Cabo Verde: 0.020
    - Mozambique: 0.090
    - Nigeria: 0.015
    - Rwanda: 0.075
    - Senegal: 0.028
    - Tanzania: 0.015
    - Uganda: 0.030
  - Note: "In this case, there are many control countries that obtain a weight smaller than 2%. For reasons of legibility, we only report the most important countries, i.e. the ones with weights larger than 2%."
- Table A3 (Synthetic controls and fit indices for CPI)
  - Fit Index by treated country:
    - Cabo Verde: 0.041
    - Mozambique: 0.104
    - Nigeria: 0.108
    - Rwanda: 0.064
    - Senegal: 0.067
    - Tanzania: 0.050
    - Uganda: 0.035
- Table A4 (Synthetic controls and fit indices for capital stock)
  - Fit Index by treated country:
    - Cabo Verde: 0.016
    - Mozambique: 0.065
    - Nigeria: 0.039
    - Rwanda: 0.075
    - Senegal: 0.023
    - Tanzania: 0.019
    - Uganda: 0.021
- Table A5 (Synthetic controls and fit indices for foreign-owned capital stock)
  - Fit Index by treated country:
    - Cabo Verde: 0.181
    - Mozambique: 0.149
    - Nigeria: 0.291
    - Rwanda: 0.153
    - Senegal: 0.078
    - Tanzania: 0.206
    - Uganda: 0.091
  - Data limitation: "Nigeria has to be disregarded in this analysis since no “fitting” synthetic control can be constructed from the donor pool (see Figure 8)."

### Comparative covariate balance between treated countries and synthetic controls (selected covariates, exact values preserved)
- Real GDP per capita covariates (selected rows; Treated / Synthetic):
  - investment rate:
    - Uganda: 33.0054 / 29.07862
    - Cabo Verde: 19.27152 / 19.54908
    - Mozambique: 18.09357 / 23.58386
    - Nigeria: 19.59305 / 18.79299
    - Rwanda: 21.99165 / 22.22125
    - Senegal: 22.19416 / 22.48265
    - Tanzania: 24.06709 / 24.09635
  - openness:
    - Uganda: 88.06105 / 83.42409
    - Cabo Verde: 62.56376 / 37.89148
    - Mozambique: 62.05576 / 62.19325
    - Nigeria: 34.35895 / 51.69207
    - Rwanda: 64.39357 / 69.30992
    - Senegal: 46.09192 / 45.91901
    - Tanzania: 33.76886 / 59.81523
  - population density:
    - Uganda: 105.529 / 98.83759
    - Cabo Verde: 22.75006 / 19.35472
    - Mozambique: 130.667 / 72.37676
    - Nigeria: 332.5216 / 48.58339
    - Rwanda: 51.05407 / 32.08761
    - Senegal: 37.69561 / 36.58756
    - Tanzania: 114.7735 / 29.47778
  - secondary school enrolment rate:
    - Uganda: 63.6288 / 39.97549
    - Cabo Verde: 8.462005 / 27.58914
    - Mozambique: 28.9348 / 37.76299
    - Nigeria: 15.85023 / 37.38375
    - Rwanda: 17.23329 / 31.1894
    - Senegal: 5.3636 / 28.93477
    - Tanzania: 14.73376 / 26.21972
  - real gdp per capita (2005):
    - Uganda: 2568.174 / 2564.414
    - Cabo Verde: 320.5381 / 302.2705
    - Mozambique: 1794.207 / 1777.037
    - Nigeria: 464.4597 / 463.1019
    - Rwanda: 750.8562 / 738.2465
    - Senegal: 444.8912 / 437.0669
    - Tanzania: 463.8207 / 455.6814
- CPI covariates (selected rows; Treated / Synthetic):
  - exchange rate regime:
    - Uganda: 2.4 / 2.3922
    - Cabo Verde: 6.066667 / 4.744067
    - Mozambique: 6 / 4.293182
    - Nigeria: 5.133333 / 5.131267
    - Rwanda: 1 / 1.318
    - Senegal: 6.866667 / 4.5462
    - Tanzania: 7 / 5.766
  - central bank independence:
    - Uganda: 0.2 / 0.19979
    - Cabo Verde: 0 / 0.0495
    - Mozambique: 0.2 / 0.183977
    - Nigeria: 0.2 / 0.199523
    - Rwanda: 0.1125 / 0.078506
    - Senegal: 0.106667 / 0.106748
    - Tanzania: 0.075 / 0.170227
  - CPI (2000):
    - Uganda: 94.92079 / 94.69103
    - Cabo Verde: 57.73669 / 57.71965
    - Mozambique: 48.21876 / 48.39501
    - Nigeria: 72.25926 / 72.20569
    - Rwanda: 92.74205 / 94.19837
    - Senegal: 80.14812 / 80.16261
    - Tanzania: 80.40698 / 81.32856
- Capital stock in percent of GDP covariates (selected rows; Treated / Synthetic):
  - openness:
    - Uganda: 88.06105 / 88.17957
    - Cabo Verde: 62.56376 / 89.31825
    - Mozambique: 62.05576 / 101.3527
    - Nigeria: 34.35895 / 117.9883
    - Rwanda: 64.39357 / 84.36941
    - Senegal: 46.09192 / 98.28676
    - Tanzania: 33.76886 / 93.70233
  - population:
    - Uganda: 425281.8 / 5408564
    - Cabo Verde: 1.79E+07 / 647959.2
    - Mozambique: 1.19E+08 / 4503795
    - Nigeria: 8193333 / 500947.5
    - Rwanda: 9837500 / 9183434
    - Senegal: 3.34E+07 / 2.70E+07
    - Tanzania: 2.29E+07 / 7785572
  - capital stock in pct of gdp (2005):
    - Uganda: 687.4423 / 695.3142
    - Cabo Verde: 339.3521 / 373.453
    - Mozambique: 437.9172 / 467.3235
    - Nigeria: 330.6231 / 348.0999
    - Rwanda: 883.7039 / 884.2657
    - Senegal: 1438.64 / 1463.466
    - Tanzania: 635.4164 / 638.6998
- Foreign-owned capital stock covariates (selected rows; Treated / Synthetic):
  - openness:
    - Uganda: 88.06105 / 88.13308
    - Cabo Verde: 62.56376 / 90.69297
    - Mozambique: 34.35895 / 91.94586
    - Nigeria: 64.39357 / 74.96442
    - Rwanda: 46.09192 / 43.99183
    - Senegal: 33.76886 / 78.77996
  - population:
    - Uganda: 425281.8 / 810335.9
    - Cabo Verde: 1.79E+07 / 3612852
    - Mozambique: 8193333 / 583074.4
    - Nigeria: 9837500 / 1575414
    - Rwanda: 3.34E+07 / 2.22E+07
    - Senegal: 2.29E+07 / 3572634
  - for. cap. stock in pct of gdp (2005):
    - Uganda: 33.04366 / 34.19779
    - Cabo Verde: 35.00573 / 33.55399
    - Mozambique: 2.990634 / 3.342479
    - Nigeria: 15.34797 / 15.56053
    - Rwanda: 26.21858 / 27.03011
    - Senegal: 18.39113 / 17.99228

### Appendix B — Effectiveness of IMF programs over time (PSM analysis, 1980-2015 and 2000-2015)
- Context and method:
  - "Applying the SCM to PSI-treated countries suggests that this type of IMF program has had a positive effect on economic development in those countries."
  - Because PSI program-type was only introduced in 2005, the Appendix analyzes IMF programs more generally over 1980-2015 using Propensity Score Matching ("PSM").
  - PSM first estimates the probability of entering an IMF program via a probit regression, then matches treated countries to controls with similar probabilities, and computes treatment effects as differences in average outcomes.
  - Variables included in the probit regression (all lagged):
    - level of real GDP per capita;
    - growth rate of real GDP per capita;
    - current account balance (as percent of GDP);
    - overall fiscal balance (as percent of GDP);
    - growth rate of terms-of-trade;
    - percentage point change in international reserves (as percent of GDP);
    - a dummy variable indicating a banking or currency crisis.
- Probit regression results (Table B1; exact coefficients, t-statistics in parentheses, significance indicated):
  - Full sample (Column 1):
    - real GDP/cap (level): −0.0000385 ∗∗∗ (−7.85)
    - real GDP/cap (growth): −0.0084595 ∗ (−1.90)
    - current account balance: −0.0092706 ∗∗ (−2.46)
    - fiscal balance: −0.0127646 ∗∗ (−2.14)
    - terms-of-trade (growth): 0.0028846 (1.54)
    - intl. reserves (change): −0.0005187 (−1.04)
    - banking or currency crisis?: 0.5713349 ∗∗∗ (3.91)
    - pseudo-R2: 0.1030
    - obs: 3,210
  - LIDC sample (Column 2):
    - real GDP/cap (level): −0.0001871 ∗∗∗ (−4.44)
    - real GDP/cap (growth): 0.0012298 (0.21)
    - current account balance: −0.0048177 (−1.04)
    - fiscal balance: −0.0161166 ∗ (−1.88)
    - terms-of-trade (growth): 0.0019092 (0.74)
    - intl. reserves (change): −0.0003565 (−0.99)
    - banking or currency crisis?: 0.4276361 ∗∗ (2.13)
    - pseudo-R2: 0.0349
    - obs: 948
  - Note from Table B1: "t-statistics in parentheses; robust standard errors are clustered at the country-level. Condition index = 4.2, suggesting multicollinearity is not a major concern in this regression. * denotes significance at the 10% level, ** implies significance at the 5% level, *** indicates significance at the 1% level."
- PSM average treatment effect on treated (LIDCs; Table B2; 5-year averages; z-statistics in parentheses):
  - 1980-2015 (Column 1):
    - real GDP/cap (growth): −1.165666 ∗∗ (−2.11)
    - inflation: 0.6269521 (0..81)
    - obs: 732
  - 2000-2015 (Column 2):
    - real GDP/cap (growth): −0.308217 (−0.45)
    - inflation: 0.2236667 (0.33)
    - obs: 732
  - Interpretation from the text:
    - Using the full 1980-2015 sample, IMF programs are associated with a negative impact on real GDP per capita growth of "about 1 percentage point per year" (statistically significant).
    - For the post-reform period 2000-2015, the negative effect on subsequent economic growth "disappears" (estimate not significant), and the inflation impact remains insignificant with the point estimate closer to zero.
    - The authors note this is consistent with the hypothesis that IMF program effectiveness in LIDCs improved after the 1999 reforms (PRGT-reforms), which emphasized poverty reduction and country ownership and relied less on conditionality.
    - The Appendix concludes: "We hope that future research will be able to address this finding in greater detail and provide evidence on the exact mechanism."

### Key analytic findings and implications (exact phrasing preserved where applicable)
- SCM for PSI-treated countries:
  - "Applying the SCM to PSI-treated countries suggests that this type of IMF program has had a positive effect on economic development in those countries."
  - This contrasts with earlier studies (Barro and Lee (2005) and Dreher (2006)) that found negative or no treatment effects for other program types; a possible reason is that those earlier studies largely analyzed data generated before the 1999 reforms to the IMF.
- Evolution over time (PSM results highlight):
  - 1980-2015: IMF programs associated with a statistically significant decrease in real GDP per capita growth (−1.165666 ∗∗).
  - 2000-2015: Negative growth effect disappears (−0.308217, not significant); inflation effects remain insignificant and move closer to zero.
  - Suggested explanation: changes from the 1999 PRGT-reforms may have improved program effectiveness in LIDCs.
- Data and methodological notes:
  - For SCM donor pools many control-country weights are below 2%; the Appendix reports only weights >2% for legibility.
  - Nigeria could not be fitted for foreign-owned capital stock synthetic control and is disregarded in that analysis.
  - PSM probit specifications include lagged macroeconomic variables and a banking/currency crisis dummy; results generally have expected signs and are more significant in the full sample than in the LIDC-only sample.

*Source: wp17109 - 2016 (IMF working paper content provided).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17109.pdf_