بحران هویت؟ بررسی برنامه های مالی صندوق بین المللی پول

The IMF uses its well-known “financial programming” model to derive monetary and fiscal programs to achieve the desired macroeconomic targets in countries undergoing crises or receiving debt relief. This paper considers under what conditions financial programming would work best, and then tests those conditions in the data. The key restrictions of financial programming are assumptions about exogeneity of some components of identities with respect to others, and the assumption of stable and “reasonable” parameters for some very simple behavioral relationships. In at least the literal applications of the framework, financial programming does not do well in forecasting the target variables, even when some components of the identity are known with certainty.

One of the most widely used applied models in macroeconomics is the financial programming (FP) model of the International Monetary Fund. The IMF utilizes the monetary, balance of payments, and fiscal identities in its design of macroeconomic programs for developing countries with goals for inflation and foreign exchange reserve accumulation, and secondarily for calculating debt relief requirements and import requirements for growth. As Barth et al. (2000) write in the official training manual for IMF financial programming, the accounting framework “is helpful in policy simulations and in analyzing the ramifications of policy options” (p. 210).1 Likewise, Blejer, Leone, Rabanal, and Schwartz (2002, p.5) note “quantitative macroeconomic performance criteria in Fund programs do not typically rely on a specific macroeconomic model. They do, however, make use of various balance-sheet identities that link monetary and fiscal variables with the balance of payments, to ensure that the Fund program is internally consistent.”2Mussa and Savastano (1999) note that a “blueprint” that contains “a preliminary assessment of the proximate and underlying sources of the aggregate imbalances” is based on “a simple flow-of-funds accounting the framework of key macroeconomic relationships.” Iteratively applied, Mussa and Savastano, 1999 say, this blueprint “enables the staff and the authorities to assess in simple quantitative terms the interactions between the policy measures agreed and the main targets of the adjustment programs.” Mussa and Savastano say the policy measures “on which almost all IMF programs focus are the public sector deficit and the creation of domestic credit by the central bank.” Of course, all macroeconomic models contain identities, and it makes no sense to “test” identities as they have to hold by definition. However, there are many different ways to use identities, and a particular use of identities may impose restrictions that are rejected by the data. How does FP use identities? The simple version of FP would recognize three types of variables in accounting identities. First, one of the elements in it is a target variable, which will absorb movements in the other components of the identity. The paper will call this the endogenous variable. Second, there is another element upon which the IMF is acting through its conditions or its own actions, such as net domestic credit or loan disbursements. The paper will call this the policy variable. The policy variable is assumed to be exogenous with respect to the target variable. Third, there are other elements in the identity that are projected exogenously or with econometric equations. The paper will call these the exogenous variables. The definition of exogeneity does not rule out their being affected by many other economic variables; typically these responses are taken into account in the projection. The exogeneity is with respect to the policy variable—they are assumed not to respond to changes in the policy variable (nor do they affect the policy variable). In other words, changes in the policy variable will affect the endogenous variable (the residual in the identity) but not the exogenous variables (the other variables in the identity). The key is that the exogenous variable is projected independently of the policy variable, based on the assumption that they are orthogonal to each other. The exogeneity restriction is the first potential problem of FP. The paper will test this exogeneity restriction by assuming that the financial programmer knows the actual value of the policy variable for the next period, but projects the exogenous variable independently (in this paper, as a random walk). The paper will compare this forecast to the naïve forecast of the endogenous variable as a random walk, and see how much FP helped when one of the variables was known with certainty. The forecast tests are not intended to evaluate FP as a method for predicting macroeconomic variables (which is not its purpose), but only to check the exogeneity restriction. Assuming the orthogonality of the exogenous variable with respect to the policy variable to be the null hypothesis, this will allow us to estimate an unbiased coefficient when we regress the endogenous variable on the policy variable with ordinary least squares. The effect of the exogenous variables on the endogenous variable will be captured by the constant term and the error term (orthogonal to the policy variable because of the exogeneity restriction). The implication of this use of identities is to assume a one for one effect of the policy variable on the endogenous variable. This paper will test this implication. The endogenous variable is typically of concern because it affects some economic outcome of concern. For example, if money is the endogenous variable, it affects inflation. If the quantity of imports is the endogenous variable, it affects growth. Usually, the relationship between the economic outcome and the endogenous variable is checked by seeing whether a behavioral parameter, such as the elasticity of imports with respect to GDP or the velocity of money, falls within a reasonable range. The reliance on such a simple behavioral relationship between the endogenous variable and economic outcomes is the second potential problem with FP. This paper will examine just how reasonable these behavioral relationships are in practice. This paper will also ask how stable and economically meaningful are the behavioral parameters—such as the import elasticity with respect to income and the velocity of money—and how accurate are forecasts based on these parameters. A third possible problem is with measurement. Although the identities hold by definition, imperfect data coming from different sources and classification problems often imply a balancing item such as “other items, net” to make the sum of the policy variable and the exogenous variable equal to the endogenous variable. Since there is even less knowledge and theory about the behavior of this balancing item, this also makes FP more problematic. The paper will examine how large these balancing items are in practice. The paper is not necessarily testing how the IMF applies financial programming in practice, since that involves many subjective judgments by IMF staff which this paper cannot model or test. IMF practitioners suggest that the application of financial programming is considerably less mechanical than the above description would indicate. IMF staff are very aware of the complex relationships among macroeconomic variables and the endogeneity of many of the key variables. They suggest that financial programming is mainly useful as a consistency check of assumptions made for different sectors: balance of payments, fiscal accounts, and monetary balance sheets. Moreover, the program is usually arrived at iteratively as parameters change. Waivers of program conditions are frequently granted when variables do not evolve as expected. Another important clarification is that this paper does not test the effectiveness of IMF conditionality, a subject on which there is already an abundant literature.3 The paper does not test whether the conditions themselves are met, nor does it test the effect of IMF programs on development outcomes in general; instead it tests whether an endogenous variable responds as predicted to variation in the policy variable subject to conditionality (regardless of whether the condition is met on that policy variable). The paper is instead stating the conditions under which FP would perform best, and then seeing how far is the data from those conditions. The results could be thought of as a guide to the limits of the most mechanical and simple uses of financial programming, as set out in published documents, making clear how much judgment will be necessary to make it workable. These issues have not escaped the attention of previous researchers. Killick (1995) criticizes financial programming on the grounds of unstable parameters and the endogeneity of other items in the identities besides the policy and target variables. Edwards as long ago as 1989 noted that financial programming “has failed to formally incorporate issues related to the inter-temporal nature of the current account, the role of risk and self-insurance in portfolio choices, the role of time consistency and precommitments in economic policy, the economics of contracts and reputation, the economics of equilibrium real exchange rates … and the theory of speculative attacks and devaluation crises, just to mention a few of the more important recent developments in international macroeconomics.” Presumably, this list of omissions has grown even larger after another 15 years of research in international macroeconomics. Indeed, one curious thing about financial programming is how unchanged it has remained over the years despite the large changes in macroeconomic theory and empirics.

This paper agrees with Agenor and Montiel, 1999 when they say: Although all of the {Bank and Fund} models to be examined have been applied frequently in policy formulation in developing nations, we shall argue that all of them are subject to limitations that constrain their usefulness for both policy guidance and analytical work as medium-term models. Among the limitations of FP pointed out in this paper are the large statistical discrepancies in all the identities, the poor performance of predictions even when an element of the identity is known in advance with certainty, the failure of econometric tests to yield a strong association between the “policy” variable and the “endogenous” variable, and the systematic instability and high variance of the “behavioral” parameters that are used as “consistency checks” on the endogenous variables with growth and inflation targets. In conclusion, the conditions under which the simplest and the most mechanical use of financial programming would work do not hold in the data. Mechanical financial programming does not appear to be a very useful guide to macroeconomic policies in developing countries. One possible improvement this paper could imply is to use econometric relationships between the policy variable and the endogenous variable, rather than relying on exogeneity restrictions in accounting identities. Or the IMF staff and those who comment on their programs may simply benefit from knowing the limitations of mechanical financial programming, implying that macroeconomic judgements based on good theory and empirics are even more important than those previously acknowledged.