## 1. Probit Estimations of the Likelihood that a Country has a Drawing Program

## Source details

**Canonical URL:** [1. Probit Estimations of the Likelihood that a Country has a Drawing Program](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp07152.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp07152.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp07152.pdf.json)

---

### I. Introduction — scope and approach
- Objective:
  - Analyze joint decisions to establish a new program and to continue drawing under an existing arrangement to better understand the evolution of IMF credit outstanding.
- Sample and frequency:
  - Quarterly frequency during 1982–2005.
  - Focus on emerging markets defined as developing, non-PRGF eligible countries.
- Definition:
  - "Drawing program" defined from approval until the last disbursement, with caveat that timing of the first disbursement is used instead of approval if the first disbursement was more than two quarters after approval or if the program was precautionary upon approval.
- Modelling choice:
  - Participation in an IMF drawing program modeled as a binary choice using probit estimations.
- Rationale:
  - Drawing programs provide a closer link to the evolution of IMF credit outstanding because they exclude precautionary programs, off-track programs, or programs with delayed first purchases.

### II. Key empirical findings
- Program prevalence and composition:
  - On a quarterly average, drawing programs were 58 percent of total programs during the period 1980–2005, and 44 percent since 2000.
  - Emerging markets accounted for more than 86 percent of IMF credit outstanding at the end of each quarter since 1982, and more than 95 percent since the Mexican crisis in 1994.
- Main determinants of drawing program participation:
  - Country-specific factors identified as most significant:
    - Net international reserves.
    - GDP growth.
  - Global factor:
    - World GDP growth is significant overall but its importance varies over time: mattered during the 1980s debt crises but not since the 1994 Mexican crisis.
  - Additional variables significant at program inception (new drawing programs):
    - External current account.
    - Inflation.
  - Interpretation:
    - External current account and inflation are more important for the decision to start a drawing program than for continuation, highlighting adverse economic conditions at inception.
- Predictive performance:
  - Out-of-sample forecast evaluation for 2004–05: model captures three to four of the five drawing programs of the period.

### III. Evolution of Fund credit and program patterns (1982–2005)
- Measurement:
  - Real IMF credit outstanding measured as nominal non-concessional IMF credit outstanding deflated by CPI inflation of major industrial countries (the countries whose currencies compose the SDR basket).
- Cyclical behavior and crisis peaks:
  - Real IMF credit outstanding exhibits cyclical behavior with peaks corresponding to major emerging market crises: 1980s debt crisis; 1994 Mexican crisis; 1997 Asian crises; most recent crises in Argentina, Turkey and Brazil.
  - Real IMF lending during the 1980s debt crisis was larger than during the 1994 Mexican crisis but just below the most recent credit outstanding peaks.
- Relationship between credit outstanding and number of drawing programs:
  - High levels of real credit outstanding do not necessarily correspond to a large number of drawing programs throughout the period; number of drawing programs is substantially lower during recent peaks despite high real lending.
  - Countries with drawing programs have borrowed more in terms of both quota and GDP in recent crises.
- Country borrowing examples:
  - Korea borrowed about 480 percent of its quota or 1.6 percent of its GDP in 1983.
  - Korea borrowed around 1700 percent of its quota or 5.2 percent of its GDP in 1998.
- Figure annotations (as presented in source):
  - Time span: 1982q1-2005q4.
  - Labels appearing with figure: 41 Programs; 31 Programs; 21 Programs; SDR 53,7 bil.; SDR 68.7 bil.; SDR 73.3 bil.

### IV. Methodological notes and sample considerations
- Exclusions and focus:
  - Supplemental Reserve Facility not included by itself because it needs to be associated with a SBA or EFF.
  - Compensatory Financing Facility and Emergency Assistance use general resources but are far less important in terms of used resources.
- Motivation for quarterly frequency:
  - Quarterly data better capture sudden changes in macroeconomic conditions than annual frequency.
- Distributional considerations:
  - IMF members’ access linked to quotas; there is a bi-modal distribution in borrowing with one cluster below access limits and another above.
- Related literature:
  - Builds on and departs from prior work by Joyce (1992), Knight and Santaella (1997), Bird and Rowlands (2001), Elekdag (2006), Ghosh et al (2007), among others, by focusing specifically on drawing periods (SBA and EFF) rather than program approvals in general.

### V. Policy-relevant implications (as presented)
- Importance of monitoring country reserves and GDP growth, and global growth conditions, for anticipating IMF liquidity needs.
- Recognition that adverse macroeconomic conditions at program inception (external current account deficits, high inflation) are predictive of new drawing programs, suggesting early-warning focus on these variables could inform IMF capacity planning.
- Quarterly monitoring enhances timely detection of shifting program demand and credit outstanding needs.

---

### 2. It includes member countries that had at least 10 percent of credit outstanding during any

### Evolution of Real Credit Outstanding (selected figures and stylized facts)
- Figure highlights:
  - A small number of countries have driven the evolution of credit outstanding, especially since the Tequila crisis.
  - Most countries had drawing programs at the same time during the 1980s; more recent peaks driven by a small and different set of countries:
    - Mexico and Russia in 1996
    - Korea, Indonesia, and to some degree Russia in 1999
    - Argentina, Brazil, and Turkey in 2003
- Figure 2 — Real Credit Outstanding (in SDR billions, December 2005 prices):
  - SDR 73.3 bil.
  - SDR 18.9 bil.
  - SDR 68.7 bil.
  - SDR 43.5 bil.
  - SDR 54.1 bil.
- Summary of stylized facts:
  - IMF credit outstanding is volatile and its peaks usually coincide with country crises.
  - In recent years most IMF credit outstanding has been driven by a few countries.
  - These patterns increase the need for a better understanding of determinants of participation in IMF drawing programs.

### Joint Decision Nature of IMF Programs
- Typical program process:
  - Member requests arrangement → IMF staff prepares blueprint → negotiation → clearance by IMF management → approval by IMF Executive Board.
- Program implementation is a joint decision process throughout life of arrangement:
  - IMF makes resources available in installments (typically quarterly) subject to performance criteria and program review.
  - After Board approval to provide financing, the country decides whether to draw.

### Determinants of Participation in IMF Drawing Programs (Model setup)
- Univariate probit specification (reduced-form of supply-demand joint decision):
  - Dependent variable: I_{ti} = 1 if country i has a drawing program at time t; 0 otherwise.
  - Explanatory vectors:
    - X: country-specific variables (lagged to avoid endogeneity)
    - Y: political and institutional variables (lagged)
    - Z: global variables
  - Lagged dependent variable included to capture inertia and unobserved persistent factors.
- Country-specific variables included (with expected sign interpretations):
  - Real GDP growth (t-1): negative expected sign.
  - CPI inflation (t-1): positive expected sign.
  - Government budget surplus (t-1): ambiguous expected sign.
  - External current account balance (t-1): negative expected sign.
  - Net international reserves (t-1), measured as months of imports: negative expected sign.
  - Real effective exchange rate (t-1): overvaluation expected to increase probability; literature shows mixed empirical signs.
- Political and institutional variables:
  - Past programs (in last two years): expected positive effect.
  - Member country quota: higher quota expected to raise probability.
  - IMF liquidity (one minus credit outstanding over industrial countries quotas): lower liquidity (higher credit outstanding to industrial quotas) expected to reduce likelihood; variable approximates Forward Commitment Capacity considerations.
- Global variables:
  - Real world interest rates (proxied by real 3-month LIBOR): higher expected to increase probability.
  - World GDP real growth: higher expected to reduce probability.
- Lag of dependent variable: captures inertia and continuation of drawing behavior.

### Determinants of Participation in New IMF Drawing Programs
- Determinants of beginning a new drawing program may differ from determinants during life of program.
- Country-specific variables (e.g., current account) likely more relevant at beginning.
- Same set of country-specific, political/institutional, and global variables used for analysis of new drawing programs.

### Data set and construction
- Coverage:
  - 59 of the 81 current developing, non-PRGF-eligible countries, quarterly, 1982–2005.
- Data construction notes:
  - 19 percent of the data were linearly interpolated from annual data where necessary.
  - Some Eastern European countries included with about a 2 year lag due to data availability.
  - 22 developing, non-PRGF countries not included (mostly Middle East countries and small island states); as a group they had at most two drawing programs in a quarter.
  - Quarters when countries had arrears are excluded since countries with arrears cannot draw.

### Empirical results — key findings from probit estimations (1982Q1–2003Q4; subsamples 1982Q1–1993Q2 and 1993Q3–2003Q4)
- Estimation details:
  - Three sample periods: 1982q1–2003Q4, 1982q1–1993q2, 1993q3–2003q4.
  - Two alternative specifications for each period; alternative specifications (models 2, 4, 6) include only countries that had at least 3 percent of credit outstanding at any quarter of the sample (major borrowing countries).
  - Regressions estimated with robust standard errors clustered by country.
  - Marginal effects reported (measured at variable means; binary variables measured for change 0→1).
- Principal summarized results:
  - Country-specific variables:
    - Net international reserves (t-1) has the expected negative sign and is significant in all specifications.
    - Country GDP real growth (t-1) is significant with expected negative sign in most specifications.
    - External current account (t-1) generally has correct sign but is rarely significant.
    - Real effective exchange rate (t-1) significant only for major borrowers and signs vary by subsample.
  - Persistence and political factors:
    - Lag of dependent variable (Drawing Program (t-1)) is highly significant with very large marginal effect.
    - Program in the last 2 years dummy is positive and significant in all specifications covering all countries.
  - Global variables:
    - World GDP real growth has expected sign and is significant during the entire sample period, especially during the debt crisis subsample.
    - Joint F-tests indicate global variables significantly different from zero during the debt crisis subsample but not in the later subperiod; since 1994 decisions to borrow driven mostly by country-specific factors.
  - Model fit:
    - Pseudo-R-squares range reported between models, indicating relatively good fit.

### Selected marginal effects and standard errors from Table 1 (marginal effect followed by standard error and significance markers where shown)
- Net International Reserves (t-1):
  - Model (1): -0.0114  0.0022***
  - Model (2): -0.0237  0.0035***
  - Model (3): -0.0249  0.0046***
  - Model (4): -0.0409  0.0124***
  - Model (5): -0.0041  0.0015**
  - Model (6): -0.0165  0.0031***
- External Current Account (t-1):
  - Model (1): -0.0015  0.0010
  - Model (2): -0.0061  0.0028**
  - Model (3): -0.0030  0.0023
  - Model (4): -0.0065  0.0080
  - Model (5): -0.0008  0.0009
  - Model (6): -0.0048  0.0032
- Country GDP real growth (t-1):
  - Model (1): -0.0052  0.0013***
  - Model (2): -0.0115  0.0033***
  - Model (3): -0.0037  0.0019**
  - Model (4): -0.0064  0.0051
  - Model (5): -0.0058  0.0013***
  - Model (6): -0.0120  0.0033***
- Country Inflation (CPI) (t-1):
  - Model (1): 0.0110  0.0149
  - Model (2): -0.0167  0.0259
  - Model (3): 0.0197  0.0205
  - Model (4): -0.0098  0.0366
  - Model (5): 0.0157  0.0143
  - Model (6): 0.0396  0.0332
- Real Effective Exchange Rate (t-1):
  - Model (1): -0.0001  0.0001
  - Model (2): -0.0005  0.0002**
  - Model (3): -0.0004  0.0003
  - Model (4): -0.0015  0.0004***
  - Model (5): 0.0002  0.0003
  - Model (6): 0.0013  0.0007*
- Government Deficit (t-1):
  - Model (1): 0.0010  0.0020
  - Model (2): 0.0004  0.0061
  - Model (3): 0.0019  0.0027
  - Model (4): 0.0040  0.0082
  - Model (5): 0.0014  0.0024
  - Model (6): 0.0031  0.0078
- Country quota:
  - Model (1): 0.0164  0.0101
  - Model (2): 0.0147  0.0182
  - Model (3): 0.0263  0.0140*
  - Model (4): 0.0550  0.0354
  - Model (5): 0.0107  0.0091
  - Model (6): -0.0169  0.0233
- IMF liquidity (t-1):
  - Model (1): 0.0001  0.0005
  - Model (2): 0.0003  0.0014
  - Model (3): -0.0015  0.0012
  - Model (4): -0.0023  0.0025
  - Model (5): 0.0002  0.0005
  - Model (6): -0.0001  0.0014
- Real LIBOR rate:
  - Model (1): 0.0017  0.0021
  - Model (2): 0.0014  0.0065
  - Model (3): 0.0024  0.0047
  - Model (4): 0.0167  0.0120
  - Model (5): -0.0029  0.0035
  - Model (6): -0.0063  0.0089
- World GDP real growth:
  - Model (1): -0.0135  0.0052**
  - Model (2): -0.0135  0.0127
  - Model (3): -0.0397  0.0103***
  - Model (4): -0.0740  0.0242***
  - Model (5): 0.0077  0.0093
  - Model (6): 0.0194  0.0289
- Program in the last 2 years:
  - Model (1): 0.0780  0.0161***
  - Model (2): 0.0515  0.0319*
  - Model (3): 0.0974  0.0277***
  - Model (4): 0.0746  0.0506
  - Model (5): 0.0525  0.0139***
  - Model (6): 0.0376  0.0387
- Drawing Program (t-1):
  - Model (1): 0.6765  0.0340***
  - Model (2): 0.7344  0.0326***
  - Model (3): 0.5928  0.0426***
  - Model (4): 0.6785  0.0387***
  - Model (5): 0.7538  0.0417***
  - Model (6): 0.8054  0.0469***
- Additional model statistics (selected):
  - Number of Observations:
    - Model (1): 4287
    - Model (2): 1655
    - Model (3): 1900
    - Model (4): 757
    - Model (5): 2387
    - Model (6): 898
  - Pseudo R-squared:
    - Model (1): 0.62
    - Model (2): 0.59
    - Model (3): 0.58
    - Model (4): 0.51
    - Model (5): 0.68
    - Model (6): 0.69
  - Wald chi2 and p-values indicate overall model significance (all reported p-values of Wald chi2 = 0.00).

### Interpretation and implications (empirical)
- Net international reserves and real GDP growth are robust country-level predictors of drawing program participation.
- Strong persistence in drawing behavior: recent drawing activity and recent programs substantially increase probability of continued drawing.
- Global cyclical conditions (world GDP growth) mattered strongly during the debt crisis subperiod (1980s), but since 1994 borrowing decisions have been driven more by country-specific factors.
- Institutional and political variables (past programs, quota) matter but show mixed levels of statistical significance across specifications.
- Real world interest rate (real LIBOR) generally has the expected sign but is often statistically insignificant in these specifications.

---

### 0.5 and 0.7.

### Determinants of Participation in New IMF Drawing Programs (summary)
- Net international reserves and country GDP real growth are leading country-specific determinants of IMF borrowing.
- External current account:
  - Significant negative expected sign for major borrower countries: the higher the current account deficit, the more likely a major borrower is to start a new drawing program.
- Inflation:
  - Evidence that inflation (CPI) is an important determinant of the beginning of a drawing program.
- Program persistence:
  - Having a program in the previous two years increases the likelihood of a successor drawing program, possibly reflecting lower political cost and greater negotiating knowledge.
- World GDP real growth:
  - Highly significant during the 1980s debt crisis, but no role after that period.

### Probit estimation results (Table 2 — marginal effects, selected)
- Sample periods: 1982Q1-2003Q4; 1982Q1-1993Q2; 1993Q3-2003Q4.
- Key marginal effect estimates and standard errors (selected ranges as reported):
  - Net International Reserves (t-1): marginal effects ranging from -0.0120 to -0.0020; standard errors include 0.0033***, 0.0007**, 0.0011*** among others.
  - External Current Account (t-1): marginal effects from -0.0030 to -0.0003; standard errors include 0.0017*, 0.0011*.
  - Country GDP real growth (t-1): marginal effects from -0.0028 to -0.0014; standard errors include 0.0004***, 0.0008***, 0.0005**, 0.0012.
  - Country Inflation (CPI) (t-1): marginal effects from 0.0027 to 0.0136; standard errors include 0.0043**, 0.0055**, 0.0042***.
  - Real Effective Exchange Rate (t-1): marginal effects around -0.0005 to 0.0000; standard errors include 0.0001*** and 0.0002.
  - Government Deficit (t-1): marginal effects around 0.0008 to 0.0091; standard errors range 0.0007 to 0.0021.
  - Country quota: marginal effects between -0.0089 and 0.0069; standard errors 0.0032 to 0.0108.
  - IMF liquidity (t-1): marginal effects between -0.0001 and 0.0002; standard errors 0.0001 to 0.0007.
  - Real LIBOR rate: marginal effects between 0.0002 and 0.0032; standard errors include 0.0007*.
  - World GDP real growth: marginal effects between -0.0189 and -0.0047; standard errors include 0.0021***, 0.0048**, 0.0035***, 0.0082**.
  - Program in the last 2 years: marginal effects from -0.0132 to 0.0191; standard errors include 0.0059*** and 0.0065***.
- Model fit and tests (examples across models):
  - Number of Observations: 4287, 1655, 1900, 757, 2387, 898 (by model/partition).
  - Pseudo R-squared: 0.12, 0.11, 0.12, 0.10, 0.11, 0.12.
  - Wald chi2 and P-value of Wald chi2: e.g., 181.82 (0.00), 122.14 (0.00), 117.05 (0.00).
  - Wald test of Global Variables and P-values: e.g., 21.85 (0.00), 7.81 (0.02), 14.25 (0.00).
  - Wald test of Institutional Variables and P-values: e.g., 1.60 (0.45), 0.37 (0.83), 0.40 (0.82).

### Robustness checks and definition refinements
- Robustness analyses:
  - Inclusion of a time trend and country-fixed effects: significance of a few variables changes, but main results remain the same.
  - Refinement of drawing program definition: excluded periods when countries did not borrow for more than three quarters within the same program; results unchanged.
  - Note: around 19 percent of the quarterly data were linearly interpolated from annual WEO data when necessary.
- Alternative modeling suggestion:
  - A Markov switching model with two stages (drawing and not-drawing) and time-varying transition probabilities could better capture transitions but entails more parameters and convergence risks.

### Goodness-of-fit and forecasting performance (selected)
- Drawing Program model (Model 1 of Table 1; cut-off = 50 percent; In-sample 1982Q1-2003Q4; Out-of-sample 2004Q1-2005Q4):
  - In-sample:
    - Percent of observations correctly called: 94
    - Percent of Drawing Programs correctly called: 86
    - Percent of No-Drawing Programs correctly called: 96
    - False predicted drawing program for total predicted drawing programs: 15
    - Probability of an actual program given a predicted drawing program: 85
    - Probability of an actual program given a predicted no-drawing program: 3
  - Out-of-sample:
    - Percent of observations correctly called: 97
    - Percent of Drawing Programs correctly called: 86
    - Percent of No-Drawing Programs correctly called: 98
    - False predicted drawing program for total predicted drawing programs: 27
- New Drawing Program model (Model 1 of Table 2; cut-off selection via loss-function):
  - Using cut-off = 3 percent (Table 4) — In-sample:
    - Percent of observations correctly called: 63
    - Percent of New Drawing Programs correctly called: 82
    - Percent of No-New Drawing Programs correctly called: 62
    - False predicted new draw. program for total predicted new draw. programs: 92
    - Probability of an actual new drawing program given a predicted new drawing program: 81
    - Probability of an actual new drawing program given a predicted no-new drawing program: 11
  - Using cut-off = 3 percent — Out-of-sample:
    - Percent of observations correctly called: 97
    - Percent of New Drawing Programs correctly called: 20
    - Percent of No-New Drawing Programs correctly called: 98
  - Using cut-off = 2 percent (optimal when using 1993Q2-2003Q4 observations; Table 5; Out-of-sample 2004Q1-2005Q4):
    - Percent of observations correctly called: 94
    - Percent of New Drawing Programs correctly called: 60
    - Percent of No-New Drawing Programs correctly called: 94
    - Out-of-sample counts: Program predicted/actual: 32/5; No-Program predicted/actual: 241/1413; False predicted new draw. prog. for total predicted new draw. Prog.: 89
    - Probability of an actual new drawing program given a predicted new drawing program: 1/1 (reported as 1 in table); given predicted no-new drawing program: 0.5
- Forecasting over longer horizon:
  - When defining the event as having a new drawing program within the year (4 quarters), the model correctly predicted all countries' programs but one.

### Policy-relevant interpretations (from these results)
- Demand for Fund credit depends on both global and country-specific factors:
  - Country-specific factors alone increase Fund credit in proportion to the size and severity of borrowing countries' needs.
  - Presence of a global factor could drive credit outstanding substantially higher; a large number of simultaneous programs at recent exceptional access levels could significantly increase credit outstanding.
- Forecast limitations:
  - The lag of the dependent variable limits real forecasting power for precise timing.
  - Out-of-sample forecasting for 2004-5 correctly predicted 3 to 4 out of 5 new drawing programs but not necessarily the largest ones, which matters for forecasting total Fund credit.

---

### Appendix A — Countries included in the study
- Argentina *
- Bahamas, The
- Bahrain, Kingdom of
- Barbados
- Belarus
- Belize
- Botswana
- Brazil *
- Bulgaria
- Chile *
- China,P.R.: Mainland *
- Colombia
- Costa Rica
- Croatia
- Czech Republic *
- Dominican Republic
- Ecuador
- El Salvador
- Estonia
- Fiji
- Gabon
- Guatemala
- Hungary *
- India *
- Indonesia *
- Jamaica
- Jordan
- Kazakhstan
- Korea *
- Latvia
- Lithuania
- Malaysia
- Malta
- Mauritius
- Mexico *
- Morocco *
- Oman
- Panama
- Paraguay
- Peru *
- Philippines *
- Poland *
- Romania *
- Russia *
- Saudi Arabia
- Seychelles
- Slovak Republic
- Slovenia
- South Africa
- Swaziland
- Thailand *
- Trinidad and Tobago
- Tunisia
- Turkey *
- Ukraine *
- Uruguay *
- Venezuela, Rep. Bol. *

- Note: Countries with an * are part of the major borrowing countries group, which includes countries that had at least 3 percent of credit outstanding during any quarter of the sample.

*Source: _wp07152 - 1. Probit Estimations of the Likelihood that a Country has a Drawing Program (excerpt).*

### 1. Probit Estimations of the Likelihood that a Country has a Drawing Program.................13

### 1. Probit Estimations of the Likelihood that a Country has a Drawing Program

### I. Introduction — scope and approach
- Objective: analyze joint decisions to establish a new program and to continue drawing under an existing arrangement to better understand the evolution of IMF credit outstanding.
- Sample and frequency:
  - Quarterly frequency during 1982–2005.
  - Focus on emerging markets defined as developing, non-PRGF eligible countries.
- Definition:
  - "Drawing program" defined from approval until the last disbursement, with caveat that timing of the first disbursement is used instead of approval if the first disbursement was more than two quarters after approval or if the program was precautionary upon approval.
- Modelling choice:
  - Participation in an IMF drawing program is modeled as a binary choice using probit estimations.
- Rationale:
  - Drawing programs provide a closer link to the evolution of IMF credit outstanding because they exclude precautionary programs, off-track programs, or programs with delayed first purchases.

### II. Key empirical findings
- Program prevalence and composition:
  - On a quarterly average, drawing programs were 58 percent of total programs during the period 1980–2005, and 44 percent since 2000.
  - Emerging markets accounted for more than 86 percent of IMF credit outstanding at the end of each quarter since 1982, and more than 95 percent since the Mexican crisis in 1994.
- Main determinants of drawing program participation:
  - Country-specific factors identified as most significant:
    - Net international reserves.
    - GDP growth.
  - Global factor:
    - World GDP growth is significant overall but its importance varies over time: mattered during the 1980s debt crises but not since the 1994 Mexican crisis.
  - Additional variables significant at program inception (new drawing programs):
    - External current account.
    - Inflation.
  - Interpretation:
    - External current account and inflation are more important for the decision to start a drawing program than for continuation, highlighting adverse economic conditions at inception.
- Predictive performance:
  - Out-of-sample forecast evaluation for 2004–05: model captures three to four of the five drawing programs of the period.

### III. Evolution of Fund credit and program patterns (1982–2005)
- Measurement:
  - Real IMF credit outstanding measured as nominal non-concessional IMF credit outstanding deflated by CPI inflation of major industrial countries (the countries whose currencies compose the SDR basket).
- Cyclical behavior and crisis peaks:
  - Real IMF credit outstanding exhibits cyclical behavior with peaks corresponding to major emerging market crises:
    - 1980s debt crisis.
    - 1994 Mexican crisis.
    - 1997 Asian crises.
    - Most recent crises in Argentina, Turkey and Brazil.
  - Real IMF lending during the 1980s debt crisis was larger than during the 1994 Mexican crisis but just below the most recent credit outstanding peaks.
- Relationship between credit outstanding and number of drawing programs:
  - High levels of real credit outstanding do not necessarily correspond to a large number of drawing programs throughout the period; number of drawing programs is substantially lower during recent peaks despite high real lending.
  - Countries with drawing programs have borrowed more in terms of both quota and GDP in recent crises.
- Country borrowing examples:
  - Korea borrowed about 480 percent of its quota or 1.6 percent of its GDP in 1983.
  - Korea borrowed around 1700 percent of its quota or 5.2 percent of its GDP in 1998.
- Figure annotations (as presented in source):
  - Time span: 1982q1-2005q4.
  - Labels appearing with figure: 41 Programs; 31 Programs; 21 Programs; SDR 53,7 bil.; SDR 68.7 bil.; SDR 73.3 bil.

### IV. Methodological notes and sample considerations
- Exclusions and focus:
  - Supplemental Reserve Facility not included by itself because it needs to be associated with a SBA or EFF.
  - Compensatory Financing Facility and Emergency Assistance are noted as using general resources but are far less important in terms of used resources.
- Motivation for quarterly frequency:
  - Quarterly data better capture sudden changes in macroeconomic conditions than annual frequency.
- Distributional considerations:
  - IMF members’ access linked to quotas; there is a bi-modal distribution in borrowing with one cluster below access limits and another above.
- Related literature:
  - Builds on and departs from prior work by Joyce (1992), Knight and Santaella (1997), Bird and Rowlands (2001), Elekdag (2006), Ghosh et al (2007), among others, by focusing specifically on drawing periods (SBA and EFF) rather than program approvals in general.

### V. Policy-relevant implications (as presented)
- Importance of monitoring country reserves and GDP growth, and global growth conditions, for anticipating IMF liquidity needs.
- Recognition that adverse macroeconomic conditions at program inception (external current account deficits, high inflation) are predictive of new drawing programs, suggesting early-warning focus on these variables could inform IMF capacity planning.
- Quarterly monitoring enhances timely detection of shifting program demand and credit outstanding needs.

*Source: _wp07152 - 1. Probit Estimations of the Likelihood that a Country has a Drawing Program (excerpt).*

### 2. It includes member countries that had at least 10 percent of credit outstanding during any

### _wp07152 - 2. It includes member countries that had at least 10 percent of credit outstanding during any

### Evolution of Real Credit Outstanding
- Figure highlights:
  - A small number of countries have driven the evolution of credit outstanding, especially since the Tequila crisis.
  - Most countries had drawing programs at the same time during the 1980s; more recent peaks driven by a small and different set of countries:
    - Mexico and Russia in 1996
    - Korea, Indonesia, and to some degree Russia in 1999
    - Argentina, Brazil, and Turkey in 2003
- Figure 2 — Real Credit Outstanding (in SDR billions, December 2005 prices):
  - SDR 73.3 bil.
  - SDR 18.9 bil.
  - SDR 68.7 bil.
  - SDR 43.5 bil.
  - SDR 54.1 bil.

### Summary of Stylized Facts
- IMF credit outstanding is volatile and its peaks usually coincide with country crises.
- In recent years most IMF credit outstanding has been driven by a few countries.
- These patterns increase the need for a better understanding of determinants of participation in IMF drawing programs.

### Joint Decision Nature of IMF Programs
- Typical program process (as described by Mussa and Savastano (1999)):
  - Member requests arrangement → IMF staff prepares blueprint → negotiation → clearance by IMF management → approval by IMF Executive Board.
- Program implementation is a joint decision process throughout life of arrangement:
  - IMF makes resources available in installments (typically quarterly) subject to performance criteria and program review.
  - After Board approval to provide financing, the country decides whether to draw.

### Determinants of Participation in IMF Drawing Programs (Model Setup)
- Univariate probit specification (reduced-form of supply-demand joint decision):
  - Dependent variable: I_{ti} = 1 if country i has a drawing program at time t; 0 otherwise.
  - Explanatory vectors:
    - X: country-specific variables (lagged to avoid endogeneity)
    - Y: political and institutional variables (lagged)
    - Z: global variables
  - Lagged dependent variable included to capture inertia and unobserved persistent factors.

- Country-specific variables included (with expected sign interpretations):
  - Real GDP growth (t-1): negative expected sign.
  - CPI inflation (t-1): positive expected sign.
  - Government budget surplus (t-1): ambiguous expected sign.
  - External current account balance (t-1): negative expected sign.
  - Net international reserves (t-1), measured as months of imports: negative expected sign.
  - Real effective exchange rate (t-1): overvaluation expected to increase probability; literature shows mixed empirical signs.

- Political and institutional variables:
  - Past programs (in last two years): expected positive effect.
  - Member country quota: higher quota expected to raise probability.
  - IMF liquidity (one minus credit outstanding over industrial countries quotas): lower liquidity (higher credit outstanding to industrial quotas) expected to reduce likelihood; variable approximates Forward Commitment Capacity considerations.

- Global variables:
  - Real world interest rates (proxied by real 3-month LIBOR): higher expected to increase probability.
  - World GDP real growth: higher expected to reduce probability.
- Lag of dependent variable: captures inertia and continuation of drawing behavior.

### Determinants of Participation in New IMF Drawing Programs
- Determinants of beginning a new drawing program may differ from determinants during life of program.
- Country-specific variables (e.g., current account) likely more relevant at beginning.
- Same set of country-specific, political/institutional, and global variables used for analysis of new drawing programs.

### Data Set
- Coverage:
  - 59 of the 81 current developing, non-PRGF-eligible countries, quarterly, 1982–2005.
  - Quarterly data allows better capture of sudden changes (reserves, exchange rates) during crises.
  - Sources: IMF International Financial Statistics (IFS) and World Economic Outlook (WEO).
- Data construction notes:
  - 19 percent of the data were linearly interpolated from annual data where necessary.
  - Some Eastern European countries included with about a 2 year lag due to data availability.
  - 22 developing, non-PRGF countries not included (mostly Middle East countries and small island states); as a group they had at most two drawing programs in a quarter.
  - Quarters when countries had arrears are excluded since countries with arrears cannot draw.

### Empirical Results — Key Findings from Probit Estimations (1982Q1–2003Q4; subsamples 1982Q1–1993Q2 and 1993Q3–2003Q4)
- Estimation details:
  - Three sample periods: 1982q1–2003Q4, 1982q1–1993q2, 1993q3–2003q4.
  - Two alternative specifications for each period; alternative specifications (models 2, 4, 6) include only countries that had at least 3 percent of credit outstanding at any quarter of the sample (major borrowing countries).
  - Regressions estimated with robust standard errors clustered by country.
  - Marginal effects reported (measured at variable means; binary variables measured for change 0→1).

- Principal summarized results:
  - Country-specific variables:
    - Net international reserves (t-1) has the expected negative sign and is significant in all specifications.
    - Country GDP real growth (t-1) is significant with expected negative sign in most specifications.
    - External current account (t-1) generally has correct sign but is rarely significant.
    - Real effective exchange rate (t-1) significant only for major borrowers and signs vary by subsample.
  - Persistence and political factors:
    - Lag of dependent variable (Drawing Program (t-1)) is highly significant with very large marginal effect.
    - Program in the last 2 years dummy is positive and significant in all specifications covering all countries.
  - Global variables:
    - World GDP real growth has expected sign and is significant during the entire sample period, especially during the debt crisis subsample.
    - Joint F-tests indicate global variables significantly different from zero during the debt crisis subsample but not in the later subperiod; since 1994 decisions to borrow driven mostly by country-specific factors.
  - Model fit:
    - Pseudo-R-squares range reported between models, indicating relatively good fit.

- Selected marginal effects and standard errors from Table 1 (marginal effect followed by standard error and significance markers where shown):
  - Net International Reserves (t-1):
    - Model (1): -0.0114  0.0022***
    - Model (2): -0.0237  0.0035***
    - Model (3): -0.0249  0.0046***
    - Model (4): -0.0409  0.0124***
    - Model (5): -0.0041  0.0015**
    - Model (6): -0.0165  0.0031***
  - External Current Account (t-1):
    - Model (1): -0.0015  0.0010
    - Model (2): -0.0061  0.0028**
    - Model (3): -0.0030  0.0023
    - Model (4): -0.0065  0.0080
    - Model (5): -0.0008  0.0009
    - Model (6): -0.0048  0.0032
  - Country GDP real growth (t-1):
    - Model (1): -0.0052  0.0013***
    - Model (2): -0.0115  0.0033***
    - Model (3): -0.0037  0.0019**
    - Model (4): -0.0064  0.0051
    - Model (5): -0.0058  0.0013***
    - Model (6): -0.0120  0.0033***
  - Country Inflation (CPI) (t-1) — marginal effects and standard errors:
    - Model (1): 0.0110  0.0149
    - Model (2): -0.0167  0.0259
    - Model (3): 0.0197  0.0205
    - Model (4): -0.0098  0.0366
    - Model (5): 0.0157  0.0143
    - Model (6): 0.0396  0.0332
  - Real Effective Exchange Rate (t-1):
    - Model (1): -0.0001  0.0001
    - Model (2): -0.0005  0.0002**
    - Model (3): -0.0004  0.0003
    - Model (4): -0.0015  0.0004***
    - Model (5): 0.0002  0.0003
    - Model (6): 0.0013  0.0007*
  - Government Deficit (t-1):
    - Model (1): 0.0010  0.0020
    - Model (2): 0.0004  0.0061
    - Model (3): 0.0019  0.0027
    - Model (4): 0.0040  0.0082
    - Model (5): 0.0014  0.0024
    - Model (6): 0.0031  0.0078
  - Country quota:
    - Model (1): 0.0164  0.0101
    - Model (2): 0.0147  0.0182
    - Model (3): 0.0263  0.0140*
    - Model (4): 0.0550  0.0354
    - Model (5): 0.0107  0.0091
    - Model (6): -0.0169  0.0233
  - IMF liquidity (t-1):
    - Model (1): 0.0001  0.0005
    - Model (2): 0.0003  0.0014
    - Model (3): -0.0015  0.0012
    - Model (4): -0.0023  0.0025
    - Model (5): 0.0002  0.0005
    - Model (6): -0.0001  0.0014
  - Real LIBOR rate:
    - Model (1): 0.0017  0.0021
    - Model (2): 0.0014  0.0065
    - Model (3): 0.0024  0.0047
    - Model (4): 0.0167  0.0120
    - Model (5): -0.0029  0.0035
    - Model (6): -0.0063  0.0089
  - World GDP real growth:
    - Model (1): -0.0135  0.0052**
    - Model (2): -0.0135  0.0127
    - Model (3): -0.0397  0.0103***
    - Model (4): -0.0740  0.0242***
    - Model (5): 0.0077  0.0093
    - Model (6): 0.0194  0.0289
  - Program in the last 2 years:
    - Model (1): 0.0780  0.0161***
    - Model (2): 0.0515  0.0319*
    - Model (3): 0.0974  0.0277***
    - Model (4): 0.0746  0.0506
    - Model (5): 0.0525  0.0139***
    - Model (6): 0.0376  0.0387
  - Drawing Program (t-1):
    - Model (1): 0.6765  0.0340***
    - Model (2): 0.7344  0.0326***
    - Model (3): 0.5928  0.0426***
    - Model (4): 0.6785  0.0387***
    - Model (5): 0.7538  0.0417***
    - Model (6): 0.8054  0.0469***

- Additional model statistics (selected):
  - Number of Observations:
    - Model (1): 4287
    - Model (2): 1655
    - Model (3): 1900
    - Model (4): 757
    - Model (5): 2387
    - Model (6): 898
  - Pseudo R-squared:
    - Model (1): 0.62
    - Model (2): 0.59
    - Model (3): 0.58
    - Model (4): 0.51
    - Model (5): 0.68
    - Model (6): 0.69
  - Wald chi2 and p-values indicate overall model significance (all reported p-values of Wald chi2 = 0.00).

### Interpretation and Implications
- Net international reserves and real GDP growth are robust country-level predictors of drawing program participation.
- Strong persistence in drawing behavior: recent drawing activity and recent programs substantially increase probability of continued drawing.
- Global cyclical conditions (world GDP growth) mattered strongly during the debt crisis subperiod (1980s), but since 1994 borrowing decisions have been driven more by country-specific factors.
- Institutional and political variables (past programs, quota) matter but show mixed levels of statistical significance across specifications.
- Real world interest rate (real LIBOR) generally has the expected sign but is often statistically insignificant in these specifications.

*Source: _wp07152 - 2. It includes member countries that had at least 10 percent of credit outstanding during any*

### 0.5 and 0.7.

### _wp07152 - 0.5 and 0.7.

### Determinants of Participation in New IMF Drawing Programs
- Net international reserves and country GDP real growth are leading country-specific determinants of IMF borrowing.
- External current account:
  - Significant negative expected sign for major borrower countries: the higher the current account deficit, the more likely a major borrower is to start a new drawing program.
- Inflation:
  - Evidence that inflation (CPI) is an important determinant of the beginning of a drawing program.
- Program persistence:
  - Having a program in the previous two years increases the likelihood of a successor drawing program, possibly reflecting lower political cost and greater negotiating knowledge.
- World GDP real growth:
  - Highly significant during the 1980s debt crisis, but no role after that period.

### Probit Estimation Results (Table 2 — Marginal Effects)
- Sample periods: 1982Q1-2003Q4; 1982Q1-1993Q2; 1993Q3-2003Q4.
- Key marginal effect estimates and standard errors (selected variables, measured at means unless binary):
  - Net International Reserves (t-1): marginal effects ranging from -0.0120 to -0.0020; standard errors include 0.0033***, 0.0007**, 0.0011*** among others.
  - External Current Account (t-1): marginal effects from -0.0030 to -0.0003; standard errors include 0.0017*, 0.0011*.
  - Country GDP real growth (t-1): marginal effects from -0.0028 to -0.0014; standard errors include 0.0004***, 0.0008***, 0.0005**, 0.0012.
  - Country Inflation (CPI) (t-1): marginal effects from 0.0027 to 0.0136; standard errors include 0.0043**, 0.0055**, 0.0042***.
  - Real Effective Exchange Rate (t-1): marginal effects around -0.0005 to 0.0000; standard errors include 0.0001*** and 0.0002.
  - Government Deficit (t-1): marginal effects around 0.0008 to 0.0091; standard errors range 0.0007 to 0.0021.
  - Country quota: marginal effects between -0.0089 and 0.0069; standard errors 0.0032 to 0.0108.
  - IMF liquidity (t-1): marginal effects between -0.0001 and 0.0002; standard errors 0.0001 to 0.0007.
  - Real LIBOR rate: marginal effects between 0.0002 and 0.0032; standard errors include 0.0007*.
  - World GDP real growth: marginal effects between -0.0189 and -0.0047; standard errors include 0.0021***, 0.0048**, 0.0035***, 0.0082**.
  - Program in the last 2 years: marginal effects from -0.0132 to 0.0191; standard errors include 0.0059*** and 0.0065***.
- Model fit and tests (examples across models):
  - Number of Observations: 4287, 1655, 1900, 757, 2387, 898 (by model/partition).
  - Pseudo R-squared: 0.12, 0.11, 0.12, 0.10, 0.11, 0.12.
  - Wald chi2 and P-value of Wald chi2: e.g., 181.82 (0.00), 122.14 (0.00), 117.05 (0.00).
  - Wald test of Global Variables and P-values: e.g., 21.85 (0.00), 7.81 (0.02), 14.25 (0.00).
  - Wald test of Institutional Variables and P-values: e.g., 1.60 (0.45), 0.37 (0.83), 0.40 (0.82).

### Robustness Checks and Definition Refinements
- Robustness analyses:
  - Inclusion of a time trend and country-fixed effects: significance of a few variables changes, but main results remain the same.
  - Refinement of drawing program definition: excluded periods when countries did not borrow for more than three quarters within the same program; results unchanged.
  - Note: around 19 percent of the quarterly data were linearly interpolated from annual WEO data when necessary.
- Alternative modeling suggestion:
  - A Markov switching model with two stages (drawing and not-drawing) and time-varying transition probabilities could better capture transitions but entails more parameters and convergence risks.

### Goodness-of-Fit and Forecasting Performance
- Drawing Program model (Model 1 of Table 1; cut-off = 50 percent; In-sample 1982Q1-2003Q4; Out-of-sample 2004Q1-2005Q4):
  - In-sample:
    - Percent of observations correctly called: 94
    - Percent of Drawing Programs correctly called: 86
    - Percent of No-Drawing Programs correctly called: 96
    - False predicted drawing program for total predicted drawing programs: 15
    - Probability of an actual program given a predicted drawing program: 85
    - Probability of an actual program given a predicted no-drawing program: 3
  - Out-of-sample:
    - Percent of observations correctly called: 97
    - Percent of Drawing Programs correctly called: 86
    - Percent of No-Drawing Programs correctly called: 98
    - False predicted drawing program for total predicted drawing programs: 27
- New Drawing Program model (Model 1 of Table 2; cut-off selection via loss-function):
  - Optimal cut-off chosen by minimizing weighted sum of false predictions and missed events.
  - Using cut-off = 3 percent (Table 4):
    - In-sample:
      - Percent of observations correctly called: 63
      - Percent of New Drawing Programs correctly called: 82
      - Percent of No-New Drawing Programs correctly called: 62
      - False predicted new draw. program for total predicted new draw. programs: 92
      - Probability of an actual new drawing program given a predicted new drawing program: 81
      - Probability of an actual new drawing program given a predicted no-new drawing program: 11
    - Out-of-sample:
      - Percent of observations correctly called: 97
      - Percent of New Drawing Programs correctly called: 20
      - Percent of No-New Drawing Programs correctly called: 98
  - Using cut-off = 2 percent (optimal when using 1993Q2-2003Q4 observations; Table 5; Out-of-sample 2004Q1-2005Q4):
    - Percent of observations correctly called: 94
    - Percent of New Drawing Programs correctly called: 60
    - Percent of No-New Drawing Programs correctly called: 94
    - Out-of-sample counts: Program predicted/actual: 32/5; No-Program predicted/actual: 241/1413; False predicted new draw. prog. for total predicted new draw. Prog.: 89
    - Probability of an actual new drawing program given a predicted new drawing program: 1/1 (reported as 1 in table); given predicted no-new drawing program: 0.5
- Forecasting over longer horizon:
  - When defining the event as having a new drawing program within the year (4 quarters), the model correctly predicted all countries' programs but one.

### Policy-Relevant Interpretations
- Demand for Fund credit depends on both global and country-specific factors:
  - Country-specific factors alone increase Fund credit in proportion to the size and severity of borrowing countries' needs.
  - Presence of a global factor could drive credit outstanding substantially higher; a large number of simultaneous programs at recent exceptional access levels could significantly increase credit outstanding.
- Forecast limitations:
  - The lag of the dependent variable limits real forecasting power for precise timing.
  - Out-of-sample forecasting for 2004-5 correctly predicted 3 to 4 out of 5 new drawing programs but not necessarily the largest ones, which matters for forecasting total Fund credit.

*Source: _wp07152 - 0.5 and 0.7.*

### Appendix A

### Appendix A

### Countries included in the study
- Argentina *
- Bahamas, The
- Bahrain, Kingdom of
- Barbados
- Belarus
- Belize
- Botswana
- Brazil *
- Bulgaria
- Chile *
- China,P.R.: Mainland *
- Colombia
- Costa Rica
- Croatia
- Czech Republic *
- Dominican Republic
- Ecuador
- El Salvador
- Estonia
- Fiji
- Gabon
- Guatemala
- Hungary *
- India *
- Indonesia *
- Jamaica
- Jordan
- Kazakhstan
- Korea *
- Latvia
- Lithuania
- Malaysia
- Malta
- Mauritius
- Mexico *
- Morocco *
- Oman
- Panama
- Paraguay
- Peru *
- Philippines *
- Poland *
- Romania *
- Russia *
- Saudi Arabia
- Seychelles
- Slovak Republic
- Slovenia
- South Africa
- Swaziland
- Thailand *
- Trinidad and Tobago
- Tunisia
- Turkey *
- Ukraine *
- Uruguay *
- Venezuela, Rep. Bol. *

- Note: Countries with an * are part of the major borrowing countries group, which includes countries that had at least 3 percent of credit outstanding during any quarter of the sample.

*Source: _wp07152 - Appendix A*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp07152.pdf_