14 Introduction to Part III

Andrew Berg, and Rafael Portillo
Published Date:
April 2018
  • ShareShare
Show Summary Details
Andrew Berg and Rafael Portillo 

In Part III we present various applications of quantitative dynamic stochastic general equilibrium models to SSA countries. Unlike the previous chapters, which provided broad guidance, the emphasis here is on specific policy questions faced by the country under study—and by extension the IMF team—and on quantitative policy guidance for central banks. Each chapter mixes theory and data, along with close attention to the broader economic context.

The chapters in Part III resulted from the collaboration between economists from the IMF African and research departments, with both serving as co-authors. Our participation in Chapters 1517 dates from our time at the development macroeconomics division (research), while Chapter 18 also stems from the division’s work agenda. Chapter 19 predates those efforts and was produced by a different division in the research department (the modelling division), but it shares the same collaborative approach. The one exception is Chapter 20, which is single-authored: it was prepared when one of us (Rafael Portillo) was part of the IMF team working on the CEMAC region.

The efforts involved in deriving and applying these models, especially those in Chapters 1517, have led to a more systematic collaboration between the IMF and central banks in SSA on the topic of analytical frameworks for policy analysis and forecasting.1 As a result, and with the support of the IMF and external consultants, several central banks have been developing and using their own variant of these models to organize their internal discussion and forecasting systems.2 This is an important part of the policy modernization efforts, as discussed in more detail in Chapter 1, and points to the synergies between research, IMF surveillance or programme work, and capacity development on the ground.

In terms of analytical approach, the models in Chapters 15, 16, 17, and 19 are semi-structural, by which we mean that they are not fully derived from first principles (micro-foundations), even though each equation has a structural interpretation, e.g., IS curve, Phillips curve, monetary policy rule, and even though the models resemble those in Part II. As these models are designed to confront the data, it is crucial that they generate plausible dynamics for the variables of interest (inflation, output, etc.). This calls for relaxing some of the restrictions implied by theory, including across equations. What is lost in theoretical elegance is gained in the ability of the model to perform quantitative policy analysis, i.e., produce reasonable conditional forecasts. This does not mean that anything goes, however; we have tried as much as possible to be guided by the structure of the more explicitly micro-founded models presented in Part II. In our view, there are benefits from working on both types of models (fully micro-founded and applied/semi-structural).

1 Applied Models for Policy Analysis and Forecasting in Kenya and Rwanda

Kenya was one of the first country cases we worked on. It had come to our attention that the authorities at the Central Bank of Kenya (CBK) were interested in building a forecasting model, as part of their transition to a more forward-looking IT-style policy regime. We worked with IMF colleagues, notably R. Armando Morales, the co-author of Chapters 15 and 16, and at the time the senior economist responsible for following Kenya, as well as directly with the staff of the CBK, over several years to assist them in these efforts.

When we were first talking with the CBK, one of the CBK officials sceptically emphasized that farmers in Kenya cared a lot more about the rains than about interest rates, capturing the important points that agriculture and especially food drive the Kenyan economy, including inflation, and supply shocks provide the main action. Our rejoinder was that the CBK has a lot more influence over the interest rate than the rain, summarizing another set of important points: that even when supply shocks provide most of the volatility, it remains the unique and critical role of the central bank to provide a ‘nominal anchor’. Monetary policy should never be the dominant driver of volatility, but through good policy it can limit volatility, avoid doing harm, and allow the flexible exchange rate to help buffer shocks.

Chapter 15 grew out of this initial set of concerns: how can we disentangle the role of, on the one hand, external and domestic supply shocks and, on the other, monetary policy and other domestic policies on inflation, and what lessons can we draw for the conduct of policy and the regime? Our period of analysis, from 2007 through 2012, was characterized by two major spikes in inflation, both driven at least in part by sharp increases in global food and commodity prices, as well as the swings in global financial market sentiment and global demand related to the global financial crisis. In this context it was difficult to assess whether the monetary policy stance was aggravating these inflationary pressures, partly because of the magnitude of the external shocks hitting the country but also because the monetary policy framework itself was somewhat opaque.3

We chose to analyse this question through the application of a small semi-structural gap model, of the type advocated by Berg et al. (2006), augmented to address the dynamics of food and non-food inflation.4 The core economics of the model are simple and involve the interaction of aggregate demand and supply, as influenced inter alia by monetary policy, in determining the deviations in output and inflation from trend. The extension raises a number of interesting complications, however. Most importantly, we pay substantial attention to the dynamics of food prices and the fact that the relative price of food seemed to be trending during this period, and also to the feedback between food price inflation and general inflation expectations. We chose to model monetary policy as a forward-looking Taylor rule. Clearly this is only a rough approximation, but it seemed like a useful benchmark through which to assess the policy stance.

One advantage of using this sort of model to analyse the data is that it is both simple enough and structural enough to be useful for regular forecasting and policy analysis in support of monetary policy decisions. Chapter 15 illustrates how to interpret the data in terms of the gaps between actual and equilibrium values, and how to decompose these gaps into the structural and policy drivers. The forward-looking Taylor provides a guide to the interest rate required to stabilize inflation. The whole apparatus facilitates communication, in that it tells an economic (as opposed to econometric) story about the direction of inflation, why the objectives may or may not have been met, and how policy is serving to steer towards the target.

When we used the model to interpret the Kenyan data, we found that excessively accommodating monetary policy had been fuelling the inflationary pressures in Kenya, in addition to the external supply shocks, and that a substantial policy tightening was required. The analysis in the chapter suggests a need to clarify the policy framework, a topic which is discussed in more detail in Chapter 5. It is worth noting that this is exactly what the central bank of Kenya did shortly after the first draft of the working paper that became this chapter (in the third quarter of 2011).

Chapter 16 returns us to a recurring obsession of this book, the role of money aggregates. As we discussed in Chapters 1 and 8, this obsession derives from the strong role that money targeting regimes have played, at least nominally, in many countries in SSA, itself partly related to the IMF’s own long and persistent tradition of using quantity-based frameworks to analyse monetary policy.5 From long experience working at the IMF on developing countries and emerging markets, we had come to have doubts about the practical utility of such quantity-based frameworks, but in our work with our IMF colleagues and member central banks, we felt the need to develop an applied framework that incorporates this approach.

Kenya provided a good case in point. Associated targets on narrow money were regularly set and announced as the guideposts for monetary policy. The simplest approach might be simply to assume, following the traditional teachings of the IMF’s ‘financial programming’ approach, that money targets are set three or six months ahead, and then monetary policy steered, e.g. through open market operations that control the quantity base money, so as to hit the targets.6 How ever, as our experience elsewhere suggested would be the case, these targets in Kenya were frequently missed by large margins. And sometimes, this seemed like it may have been for good reason, for example because shifts in demand for money implied that hitting the target would have implied an unnecessary and harmful shift in the level of interest rates.

To capture money targeting as implemented in practice, Chapter 16 builds on the model in Chapter 15 by introducing some of the features we analysed in Chapter 8. These include: (i) a money demand equation, (ii) a rule specifying the setting of money targets, and (iii) a more general monetary policy setting that places some weight on hitting money targets and some weight on a more standard Taylor rule. On the latter we assess the actual weights placed on these two policy aspects and ask about the implications for macro dynamics of changing them.

We find that the CBK set money targets as if they were trying to forecast future money demand, and indeed they seem to have some ability to predict this money demand better than a very simple statistical model, implying that they use information about future liquidity shocks, e.g. due to fiscal operations.

However, we find no influence of the monetary aggregates on the stance of policy. Misses were large, and the subsequent adjustment came from shifting the target, not bringing the actual money stock back in line or from adjusting interest rates in response to target misses. Asking about the counterfactual, we see little role for greater adherence to the money targets, on the whole. The question is not simple. Echoing some of the results in Chapter 8, we find here too that greater adherence to money targets could have helped buffer demand shocks. Overall, though, more volatility would have resulted. This underscores an important general point we have made in Chapter 1, which is that money targeting is not a particularly useful solution to many of the particular challenges of monetary policy implementation in SSA, such as the dominant role of supply shocks.

We had hoped that the sort of framework developed in Chapter 16 would be useful in countries that, unlike Kenya, are interested in continuing to put some weight on monetary aggregates. We had and to some extent still continue to hope such frameworks will allow policymakers to disentangle money misses into shocks due to money demand (given adherence to an appropriate policy reaction function) and policy shocks (perhaps errors or deviations from conditionality under an IMF programme). And we and our IMF colleagues continue to work with a few such central banks along these lines. However, we are increasingly pessimistic about how fruitful this effort will be. The model and analysis in Chapter 16 is quite complex. Both in the chapter and in related operational work, we have found that the addition of one more structural equation (money demand) to a four-equation model increases the complexity involved in interpreting the data by much more than 25 per cent. As in Chapter 16, we have not found much use for this additional complexity in practice. Partly for this reason we have tended to encourage countries to simplify their frameworks and focus directly on interest rates, inflation, and the other elements of the analysis in Chapter 15.

Chapter 17 considers a country with an even more complex regime. Rwanda in practice has followed a hybrid approach. As in the stylized view of Kenya in Chapter 16, it puts some weight on money aggregates and also on an interest rate that responds to inflation, output, and other factors. At the time the chapter was written, the framework aimed at stabilization inflation, although the stabilization of the exchange rate was also central. Moreover, monetary policy operations were conducted in such a way that the ‘policy rate’—the short-term interest rate assigned an explicit role as indicator of the stance of policy—often deviated from the short-term market rate. This deviation is a common feature of regimes in which multiple objectives and complex policy frameworks are hard for the central bank to reconcile.

In Chapter 17 the authors describe this hybrid regime, but then, to keep things tractable, focus their modelling efforts in particular on a simple way to capture the fact that the authorities directly manage the exchange rate. They do so by introducing an exchange rate target in the equation that describes exchange rate dynamics—the uncovered interest parity condition—and allowing the exchange rate to be driven both by expectations of future interest rate differentials/risk premia and by the exchange rate target. Although not modelled explicitly as such, the latter mechanism is meant to capture direct management of the exchange through exchange rate interventions, in the context of limited de facto capital mobility.7 The authors argue that even with such a complex framework their simple model performs fairly well and can play a useful role in understanding and guiding policy.

The regime in Rwanda cannot be found in any textbook. And a model that captured it more fully, considering for example the role of money targets as well as the exchange rate in the objective function, might well be too opaque to be useful for regular policy analysis. However, Rwanda has had good success in coping with the food and fuel price shocks of 2008–12 (Chapter 5). This success may in part be due to other features of the overall policy regime, such as relatively flexible and countercyclical fiscal policy, but it reminds us that the mapping from policy regime to outcomes is not straightforward.

Readers steeped in the DSGE tradition may find the models in Chapters 1517 too simple and ad-hoc. However, they are meant to be brought to the data and used, with a healthy dose of judgement, to inform policy decisions on a quarterly basis. This is in many ways a much more demanding task than building a model from first principles, estimating or calibrating it, and examining its properties in terms of impulse responses (for example). Sometimes, however, more complexity is required for the question at hand, as we see in the next three chapters.

2 The Impact of The Global Financial Crisis on Zambia and the Policy Response

In Chapter 18 we began with a seemingly simple question: why are changes in policy rates often not followed by changes in lending and deposit rates? This is a commonly observed phenomenon, in SSA and elsewhere. It surfaced starkly in Zambia in 2008, when short-term rates fell by over 1,200 basis points, while the lending rate fell by about half that amount policymakers naturally wanted to know why. One set of views was that long rates are simply sticky or, for one reason or another, are not market-determined. We preferred to try to understand the situation explicitly in terms of economic forces.

This narrow question provided a window into a much broader question that forms the core of the chapter: how should we think about the impact of the global financial crisis on low-income countries, and in particular how should we model the most important channels. We explicitly modelled the banking system, because it is the banking system that translates short-term interbank or policy rates into lending and deposit rates, and more generally because we viewed it as a key source of transmission of the crisis. We viewed the crisis, for Zambia, in terms of three related shocks: a decline in the terms of trade, an increase in the country’s external risk premium as part of a general ‘flight to quality’, and a decrease in the risk appetite of Zambian banks in the face of a general crisis-related increase in risk aversion.

Our preliminary hypothesis was that a standard ‘financial accelerator’ mechanism, a workhorse of macroeconomic models with banks, could help explain the trajectory of spreads. The idea is that the risk premium that banks demand on their loans relative to their cost of funds will depend on the stock of loans relative to the value of collateral, which is in part measured by stock market valuations or perhaps simply the value of output. In Zambia, according to this mechanism, the drastic decline in the price of Zambia’s dominant export, copper, in 2007/08, which lowered asset prices and the value of collateral in Zambian firms, was responsible for raising the risk premium on loans.

A closer look at the data, through the lens of the DSGE model in Chapter 18, showed us the limits of this view for the episode in question. Simply put, the collapse in lending was much greater than the fall in asset prices, output, or other available measures of the value of collateral. So something besides an increase in the ratio of loans to collateral is needed to explain the rise in spreads.

We came to conclude that a shock to the risk aversion of the banks themselves was the best way to explain the data. In the face of the confusion and uncertainty of the global financial crisis, banks in Zambia switched preferences strongly towards investing in relatively safe government securities and deposits at the central bank, reducing as much as possible their exposure to firms. At the same time, foreign lenders attempted to reduce exposure to Zambia. The result was a sharp decline in lending, emergence of a large current account surplus, a rise in demand for liquid assets, and a rise in the spread between policy and lending rates.

Our use of a structural model allows us to ask about the role of the monetary policy regime. We find that some features of the regime may have led to a ‘stop and go’ policy that may have increased the real effects of the global financial crisis. The ‘stop’ was in increase in policy rates of some 400 basis points through mid-2009, which we attribute to a (backward-looking) concern about inflation, a desire to resist the large exchange rate depreciation associated with the negative terms of trade shock, and a concern about the large increase in ‘liquidity’ resulting from the risk-induced shift of banks from loans to high-powered money. A more forward-looking regime without the emphasis on monetary aggregates might have avoided this ‘stop’ and buffered the shocks a bit better. We should not overstate the magnitude of this effect, however: we assess that most of the effect of the crisis was due to real external shocks, and monetary policy could do relatively little to help. This again illustrates a general lesson, which is that we should not expect too much from monetary policy. It cannot affect trend relative prices (Chapter 15), and it cannot fully mitigate the effects of large real shocks.

3 Endogenous Credibility and the Cost of Disinflation in Ghana

Chapter 19 brings into sharp relief some major weaknesses of all the models in the rest of this book, while providing an alternative. For good reason, we have relied entirely on so-called ‘rational expectations’ models, where the central bank has complete credibility. But this feature creates some major pitfalls in the application of these models.

Rational expectations, in which agents’ expectations of future variables are also the model’s own predictions, represent a simplification. Purely backward-looking models are simpler still, but they ignore the critical role that monetary policy regimes play in anchoring expectations. The lessons of the great inflation of the 1980s in advanced countries, and the resulting rational expectations revolution in macroeconomics, apply a fortiori to countries in SSA: people are reasonably quick to adapt their behaviour to the nature of the regime, and policies that attempt to systematically fool the public are not likely to be sustainable.

A dangerous feature of the rational expectations models we have been using so far is that any deviation of policy from that dictated by the policy rule in the model is assumed by agents in the model to be a one-off. This means that these models are not well-suited to answer one sort of question a central bank governor might ask: ‘Suppose I lower policy relative to what we normally would do—what would happen?’. The problem is that the resulting boost to inflation is small and, in the model, not a problem, because agents assume that the deviation is certainly temporary and not symptomatic of a lack of willingness of the central bank to do what is necessary to achieve its objective. The perfect credibility of the central bank—the confidence that it will return to its policy rule and bring inflation back down—mitigates the inflationary consequences. Similarly, a decision to delay inflation stabilization because of an aversion to the policy tightening that would be called for by the announced regime would generally only have temporary and relatively minor effects.

In real life, though, the demand contraction required to reduce inflation back to target may be larger than the boom that generated that inflation, partly because the unexpected loosening may raise doubts about the goals of the central bank, requiring a larger contraction to convince the public that inflation will indeed be brought back to target.

In Chapter 19, the authors apply to Ghana a richer version of the standard framework we have seen several times, a version that introduces the notion that the credibility of the central bank depends on its inflation record. They apply this framework to discuss the question of the appropriate pace of disinflation from a position of imperfect credibility, arguing that appropriate front-loading of the required contraction will lower the total cost in terms of output.

The simpler models presented in earlier chapters remain the workhorse models in many central banks, particularly as they begin using policy-based policy analysis. And the assumption of perfect credibility remains a useful benchmark; the evidence suggests that credibility is generally gradually established as central banks implement inflation targeting regimes, and a central bank that behaves as it should under perfect credibility is in effect leading by example. However, the lessons of Chapter 19 need to be kept in mind. The analytic approach in this chapter has proven useful in other contexts (e.g. Israel in Argov et al., 2007, and India in Benes et al., 2017). And the chapter illustrates how more complex models can guide the use of simpler core models. In practice, the simpler models can be ‘tuned’ in an ad hoc fashion to capture the implications of weak credibility in particular cases.

4 The Role of External Shocks and Fiscal Policy in the Dynamics of Inflation in the CEMAC Region

Chapter 20 takes us in another direction entirely. In almost the entire book, we have focused on countries with some degree of exchange rate flexibility, a natural approach given the scope this flexibility provides for monetary policy. However, about half the countries in SSA have a hard exchange rate peg, most of them as members of currency unions that have maintained a fixed parity with the French franc/euro (with a large step devaluation in 1994). Chapter 20 analyses the determinants of inflation in the six countries that make up the Communauté Economique et Monetaire de I’Afrique Centrale (CEMAC).

In principle and, in our experience in practice, countries with a peg but with some degree of closure of the capital account—which could come from legal restrictions or the imperfect substitutability of their assets for those of other countries—do face some choice with respect to the stance of monetary policy. However, the VAR-based empirical analysis in the chapter points to the importance of fiscal shocks, combined with a passive monetary policy, along with the usual supply shocks. Counterfactual analysis with a calibrated structural model that is consistent with this VAR evidence points to the merits of this passive monetary policy. A more aggressive anti-inflationary monetary policy may be feasible, but it would choke off the real exchange rate adjustment that helps adjust to shocks and thus likely increase output volatility.8


    ArgovE.EpsteinN.KaramP.LaxtonD. andRoseD. (2007). Endogenous Monetary Policy Credibility in a Small Macro Model of Israel. IMF Working Paper 07/207. Washington, DC: International Monetary Fund.

    • Search Google Scholar
    • Export Citation

    BenesJ.ClintonK.AsishG.GuptaP.JohnJ.KamenikO.ZhangF. (2017). Quarterly Projection Model for India: Key Elements and Properties. IMF Working Paper 17/33. Washington, DC: International Monetary Fund.

    • Search Google Scholar
    • Export Citation

    BergA.KaramP.D. andLaxtonD. (2006). A Practical Model-Based Approach to Monetary Policy Analysis-Overview. IMF Working Paper 06/80. Washington, DC: International Monetary Fund.

    • Search Google Scholar
    • Export Citation

    PolakJ. J. (2005). The IMF monetary model at forty. In J.Boughton (Ed.) Selected Essays of Jacques J. Polak 1994–2004. New York and London: Sharpe20926.

    • Search Google Scholar
    • Export Citation

Financial support by the UK’s Department for International Development (DFID) has been pivotal in these efforts.


This includes the central banks in Ghana, Kenya, Mozambique, Rwanda, Tanzania, and Uganda.


Chapter 5 takes a complementary narrative look at Kenya over the same period, comparing to Uganda, Rwanda (also discussed in Chapter 17), and Tanzania.


The structure of these models is similar to those in Chapters 8 and 9, though as mentioned above it breaks some of the restrictions implied by theory, mainly by allowing for ad-hoc backward-looking dynamics.


See Polak (2005), among others.


‘Base money’ is the money that is directly created by the central bank, and consists of deposits of commercial banks at the central banks as well as cash in circulation. ‘Broad money’ includes in addition checking and savings deposits at commercial banks.


Chapter 13 provides a more formal treatment of this mechanism.


Chapter 20 was written years before Chapter 6, which identified the challenges to VAR-based analysis in low-income countries. However, unlike almost all flexible-exchange-rate countries, CEMAC countries enjoy relatively long time series under a stable policy regime. Moreover, the chapter does not attempt to identify monetary policy shocks in the VAR, employing a DSGE model to help articulate the role of the monetary policy regime in propagating other, real, shocks.

    Other Resources Citing This Publication