Journal of Policy Modeling
Subscription Information Search About This Page Back to the Homepage

Back to Current Issues

Financial Liberalisation and Price Stability in Kenya[1]

Anders Isaksson

Department of Economics, Göteborg University Box 640 S-405 30, Göteborg, Sweden.


It has been postulated in the literature that attempts to liberalise the financial sector when inflation is high can lead to high interest rates and even higher inflation. Thereafter. when inflation is fought, a period of low inflation and high real interest rates follow. Since Kenya experienced this sequence, it appears that prices were unstable before and during the financial liberalisation. This paper argues that this was not the case as evidenced by cointegration between the involved variables and the abliltv to estimate a stable inflation model over the period 1970-91. When the cointegrating relationship breaks down, which it does in Kenya after the financial liberalisation, economic agents can no longer forecast inflation with confidence using Historical data. This breakdown of the cointegratina vector implies that agents switch to forward-looking behaviour, perhaps an indication of lack of credibility, in the financial liberalisation process.

JEL Classification numbers: E31, E44 and O11

Kevwords: Kenya, Price stability, Financial liberalisation, Cointegration, VAR.

1. Introduction

As part of its financial-sector reform, Kenya liberalised interest rates between January 1988 and July 1991[2]. Subsequently, market interest rates skyrocketed, while inflation rose even further. When undertaking financial liberalisation under conditions of high and unpredictable inflation, interest rates might rise in order to offset anticipated inflation and to balance supply and demand for loanable funds (McKinnon, 1988, 1991). Rising domestic interest rates may lead to large capital inflows that in turn cause inflation if not sterilised. High real interest rates also reduce borrower net worth, which has a negative impact on investment and financial intermediation, leading to rising non- performing assets and bank failures. Under such circumstances implementing a financial liberalisation is difficult. Kenyan inflation reached unprecedented levels in 1992-93, forcing the government to attempt to halt it by pursuing restrictive monetary policy. As the economy stagnated at the same time, the prediction of the neo-structuralists that financial liberalisation may lead to stagflation seemed fulfilled[3]. However, while the Government had managed to control inflation by 1994, real interest rates remained high, indicating a continued high cost of investment. In these circumstances it is high-return, high-risk projects that are financed suggesting instances of adverse selection and moral hazard (Stiglitz and Weiss, 1981).

The main purpose of this paper is to study price stability in Kenya prior to and during financial liberalisation. Price stability will be defined as a situation where economic agents with some accuracy can forecast future inflation. This presumes that they have a fair knowledge of the causes of inflation and the magnitude with which individual variables affect it. This knowledge reduces future uncertainty creating a more favourable environment for economic growth. The basic assumption in this paper is therefore that cointegration between the price level and its determinants depicts such a situation.

The rest of the paper proceeds as follows: section 2 provides an overview of Kenya's financial-sector reform and also discusses the prevailing macroeconomic situation. Section 3 presents an inflation model for Kenya. In this section an examination of the time-series characteristics of the individual variables and structural breaks in the consumer price series are examined in order to obtain some knowledge regarding price stability. In section 4 tests for cointegration are performed and the implications for price stability are discussed. Section 5 evaluates the estimation results of the inflation model, while section 6 concludes the paper.

2. Kenya's Financial-Sector Reform: An Overview

The financial-sector reform started in 1989, when assistance was received from the IN4F and the World Bank, although the government had already started implementing some reforms by January 1988. This reform included policy and institutional measures. Policy reform included interest-rate liberalisation, development of money and capital markets, improvement of the efficiency of financial intermediation, development of more flexible monetary-policy instruments, removal of credit ceilings, and reduction of both the government's excessive reliance on domestic bank borrowing and reduction of its budget deficit. Institutional reforms were aimed at setting up a regulatory framework, and ensuring prudential regulation and supervision of the financial system. Further, there was emphasis on the need to restructure the troubled financial institutions, including privatisation, and improvement of technical expertise at the Central Bank of Kenya.

The first donor-supported financial sector reform lasted up to 199 1. but was soon followed by further reforms initiated by the government of Kenya itself. They included measures to ensure current and capital account convertibility, notablv removal of controls on foreign exchange transactions. The Central Bank increasingly undertook open market operations, improved reserve money management and regulation of the banking system.

To provide some indication of the pressures that led to stabilisation and structural adjustment, including financial-sector reform, a macroeconomic overview covering the period 1980-1995 of the Kenyan economy is undertaken. The discussion is based on selected indicators presented in Table 1.

During 1980-85, and prior to the second structural adjustment programme[4], annual inflation averaged 13%, while the real interest rate was negative. Fiscal and current account deficits were both above 5% of GDP. Growth of money (M 1), was, however, low, only 6.2%, while domestic: credit of the banking sector was 46% of GDP. Real economic growth averaged 3%, that is, barely at par with the p pulation growth rate. In the three-year period (1986-88) before the financial liberalisation, inflation was about 10% on average, while real interest rates turned positive. The fiscal deficit decreased to about 5% of GDP and the current account deficit decreased also to 4.2% of GDP. Real economic growth rose to an average of 6.5%. Money growth increased by about 15%, partly driven by increased domestic credit and the depreciation of the shilling against the US dollar.

Table I Kenya: Selected Indicators, 1980-95.

Table 1

Conditions seemed stable enough to permit introduction of financial liberalisation, but when it began in 1989, inflation had risen to 13.6% and the real interest rate was failing. Although growth dropped to 4.2% this was still hi-her than in the early 1980s. The currency depreciated with inflation, while money growth decreased a little and domestic credit fell back to 47% of GDP. In order to broaden and reinforce the adjustment process, the Structural Adjustment Facility (SAF) arrangement was replaced in May 1989 by a three-year Enhanced Structural Adjustment Facility (ESAF). Among the objectives was real per capita growth of GDP, reducing inflation and the current-account deficit, and to build up Kenya's net official international reserves. However, in 1990, inflation continued to rise as a result of rising excess private demand, reaching 20% and the real interest rate reached minus 5%.

Perhaps the most disruptive event in Kenya's stabilisation efforts in the 1990s was of an external and collective nature. Wishing to influence Kenyan policy with regard to transparency and accountability in the use of funds, official corruption, human rights, and multiparty democracy, the donor community decided to withhold aid. Between the fiscal years 1990 and 1991, net external loan accruals fell from four billion to 230 million shilling. Via introduction of austerity measures the government managed to balance the fiscal account by 1992. However, inflation rose to 34%, partly due to a drought that increased food prices (Levin, 1994). GDP grew by only 0.3%.

The second shock, this time of a domestic nature, arose out of pressures on the ruling party to win the multiparty election in 1992. A large amount of money was released, much of it simply borrowed from the central bank. Thus in 1992 money grew by 47%. Total domestic credit at 52% of GDP remained high. Given apparent lack of policy improvement, the World Bank postponed disbursements for ongoing operations in September 1992. Discussions between the Kenyan government and the IMF resulted in a 'shadow programme', this to establish a stable macroeconomic environment. However, other donors remained reluctant to join any prospective funding. By 1993 inflation had reached close to 55% and the shilling was trading at 68 to the US dollar. In spite of the floating of the shilling, ahead of a visit by the IMF, the outcome of the meeting was unimpressive. Kenya withdrew the float on the IMF's departure. Real GDP growth continued to stagnate in 1993, which is partly explained by the aid shock that reduced private investments, cut down on resources, and caused a rationing of foreign exchange. The current account went into surplus, while the fiscal deficit started rising again. In April 1993, the Government declared its commitment to taking corrective policy measures and World Bank funds came back on stream.

In 1994, there was some evidence of stabilisation. Due to large capital inflows following liberalisation of foreign exchange transactions and high interest rates, the shilling appreciated to 45 to the dollar. The coffee boom this year also contributed to this development of the shilling. The government's deliberate efforts to mop up excess liquidity, using open market operations, brought down the money growth rate to 12.6 Domestic credit, however, increased to 52% of GDP. Inflation decreased to 6.6 which was partly due to the return of price controls. The fiscal and current account balances remained at about the same level as in 1993, and economic growth rose to 2.7%. But average real interest rates were over 17%.

In 1995 inflation was about 7%, but the real interest rate remained high at 13,7 Following a successful consultative meeting in Paris in July, which alleviated the fear of another aid embargo, the shilling stabilised. Money growth decreased to 3.8%, while domestic credit continued to increase, reaching 55% of GDP. Kenya thus emerged from a turbulent inflationary period with high positive real interest rates, a likely outcome of financial reforms, notably interest-rate liberalisation, undertaken during a period of unstable prices and overall macroeconomic instability.

3. Modelling Inflation in Kenya

This section introduces the theoretical model underlying the econometric modelling and defines price stability more explicitly.

The model presented here is an extension of Bruno and Sachs (1985) and Bruno (1993). One of its appealing features is that it incorporates both demand-pull and cost-push ingredients. This is important when analysing inflation in Kenya (for demand side arguments see Ndung'u, 1993; Mwega, 1990, and for cost-side arguments, see Sharpley and Lewis, 1990).

The model starts with the aggregate supply (Ys) and demand (Yd) for goods:

Equation 1    (1)

Equation 2    (2)

where P is the domestic price level, W is the nominal wage level, E is an effective exchange rate index, Pm is an import price index and Px is an export price index, both exogenous and in foreign currency, and M is the money supply.

The aggregate supply function is derived from a three-factor production function containing labour, capital and intermediate inputs. The marginal products of all inputs are assumed to be equal to their respective real market prices. Although it would be desirable to include cost of capital as a determinant of inflation, it has not been possible. First, and most important, there is no direct measure available. Second, although the Treasury bill rate is sometimes used as a proxy, the variable is probably inappropriate in an economy still under controls. Third, most capital in Kenya's imported and part of the cost of capital is therefore reflected in the import price index. Intermediate inputs are assumed to be separable from the value added output. Therefore, the only interesting price of intermediates is that of the imported ones. This price is approximated by that of the import-price index.

Aggregate demand is derived from an open economy IS-LM model. It is a combination of money-market and Keynesian aggregate commodity-demand equations. This demand schedule consists of consumption, exports, and investments with a substitution for the interest rate.

In logs, the supply-side of the model can be written as

Equation 3    (3)

and that for ag-reaate demand as

Equation 4    (4)

It is assumed that inflation is generated by the movement from one equilibrium to another in the goods market. Equilibrating the goods market, assuming one international price p* for both exports and imports, and long-run linear homogeneitv, i.e. b1 + b2 + b3 + b4 = 1, domestic prices can be expressed as follows

Equation 5    (5)

where g0 = a + m, g1 = b1, g2 = (b2 + b4), g3 = b3, and g4 = (b2 + b4).

Regarding economic agents, equation (5) provides the framework for their forecasts of inflation. As long as equation (5) exhibits stable parameters, economic agents can reasonably forecast future inflation even in the event of a large devaluation or rapid increase in money supply. Such a case would be characterised by price stability in this paper's meaning of the word. In other words, variables in equation (5) should be cointegrated, which is a testable hypothesis. However, a structural chance, such as that caused by structural adjustment programmes or financial-sector reform, could cause a break in the cointegrating relationship. This would last until a new (long-run) equilibrium is attained, or equivalently, until economic agents have found a new framework for their inflation forecasts.

3.1. Time Series Characteristics of the Data[7]

In this section I investigate the non-stationary properties of the data by testing for unit roots. I also try to establish the existence of a stable period by examining more closely the price variable. Recursive estimation of the price variable series may reveal appropriate break point between a stable and unstable period in the sample[8].

Domestic prices, p, are the weighted December-to-December consumer-price index (CPI). The money measure, m, is Ml, while the wage variable, w, is defined as the average wage for formal sector employees[9]. The international price index, p*, is a weighted average of export and import price indices. Calculation of the nominal effective exchange rate index, e, is based on Kenva's 10 most important trading partners[10].

Figure 1: Recursive Estimation of DF-test for Consumer Prices, 1970-95.

Figure 2

The first task is to examine whether the consumer-price series has break points that can he used to discriminate between stability and instability. This provides a first indication of whether consumer prices have been stable over the whole sample, or whether there are regime shifts (for instance, from stability to instability) that after it. One natural candidate to use for this examination is the fact that financial liberalisation occurred mainly between 1989-91. Since the interest is in stability of the period that includes the financial reform, a suggested break-point would be in 1992. To find some empirical support for 1992 as a break-point between stability and instability, the DF-test for the consumer price series is run recursively.

Figure 1 shows the one-step Chow-test for unexplained variation, obtained from this recursive estimation. Points outside the horizontal dashed line, which is the 5% significant level, are either outliers or associated with parameter changes. The plot in Figure I suggests a single break point in 1992. In other words, it indicates that inflation was stable until 1991 and unstable during 1992-94, Years that proceeded interest-rate liberalisation. Since inflation came down to 6.6% in 1994, remaining almost the same in 1995, it is possible that full sample integration tests might fail to reveal the instability shown by Figure 1.

Table 2: Unit-Root Tests, 1970-91.

Table 2

The next step is to test for unit roots in all variables with sample periods based on the recursive estimation done above. Three tests for unit roots are used, the Dickey- Fuller-test (DF), the Augmented Dickey-Fuller-test (ADF), and the Sargan-Bhargawa-test (SBDW). Tables 2 and 3 present the unit-root tests for the alleged stable and unstable periods. For the former period, all variables appears to be I(1) in levels[11]. Further evidence on the issue of stationarity of the data will be provided in section 4.

The price instability of the post-financial liberalisation period shown in Figure I is not obvious from results of Table 3[12]. The accelerating inflation part of the consumer price series was short-lived, covering only the first few years of the 1990s. When 1994- 95 data are included, the consumer price series can no longer be viewed as an I(2) process, but probably as an I(1) process with a structural break. The exchange rate appears to be 1(2). This is not surprising, since the Kenyan government pursued an active depreciation policy from the late 1980s to 1994. However, the exchange rate appreciated sharply in 1994 failing back to a much lower level and for the full sample period it is not clear whether the variable is I(2) or I(1) with a structural break.

Table 3: Unit-Root Tests, 1970-95.

Table 3

4. Testing for Cointegration

This section starts with an introduction to the econometric framework and proceeds with tests for cointegration. It also analyses the long-run relations. Imposing restrictions (based on economic theory) on the cointelyratincr vectors does this. To describe the unknown data-generating process, a vector-autoregressive (VAR) model with Gaussian errors, that is, errors that are independently and identically distributed (i.i.d), is formulated

Equation 6    (6)

where yt, is a p x l vector of stochastic variables (pt, et, pt*, wt, mt), yt+k+1......y0 are fixed, e1....... eT are NIIDp (0, S), with covariance matrix S,m is a vector of constants, and Dt are deterministic dummy variables.

When reparameterising in the error-correction form the effects of the short- and long-run variation in the data can be captured as follows

Equation 7    (7)

The model specification (7) is a general description of a generally formulated reduced-form model where the short-run dynamics, G 1….G k-1 are likely to be over- parameterised.

Depending on the cointegrating rank, r = rank (P ), the model has three possible interpretations. When r = 0, (7) is a correctly specified VAR in first differences. If r = p, (6) is a correctly specified VAR in levels, and if 0 < r < p, there are r cointegrating vectors and (7) is a vector error-correction model.

Writing, P = ab' where a and b are full rank (p x r)-matrices, the estimation of (7) involves only stationary processes. Since b'yt is stationary, the relations could be interpreted as economic long-run relations. The linear weights must, however, be economically interpretable parameters. Since the weights do not necessarily represent meaningful economic relations, testing of structural hypotheses is crucial especially when there is more than one cointegrating vector.

4.1. Idetitification of the Cointegrating Vector

The purpose of this sub-section is to identify one or more stable long-run relations in the data by means of cointegration. To start with, the error terms must be Gaussian processes. This can be ensured by including an appropriate number of lags and dummy variables. Given Gaussian errors, the proper approach is to estimate the cointegrating space first and to then test specific hypotheses of economic interest within the cointegrating space.

Tables 4A and 4B report the tests for the appropriate lag length and normality of the variables in a multivariate framework, while Table 4C presents results from univariate tests of the residuals. All Tables include models with and without dummy variables. Since data are annual, and number of observations few, only two lags were imposed. The results in Tables 4A-C suggest that only one lag was necessary and that dummy variables, which take account of the first oil crisis (DI974), the rapid recovery that started with the coffee boom (DI975) and its peak (DI977), were necessary to ensure well-behaved residuals. Having established the appropriate lag-order for Gaussian errors, the next step is to test for cointegration and to identify economically interesting long-run relations.

Table 4A: F-test for Determination of Lag-Order, 1970-91.

Table 4a

Table 4B: Multivariate Determination of Lag-Order, 1970-91.

Table 4b

Table 4C: Univariate Testing of the Residuals, 1970-91.

table 4c

Next, the cointegrating rank of Pab is investigated. An intuitive test for cointegrating rank r is the evaluation of the number of eigenvalues of P that are close to zero. Johansen (1988, 1991) derives two tests for testing this, the trace-test, trace = -TSln(l-li), and the maximum-eigenvalue test, max EV = -TSln(l-li), where li is the eigenvalue associated with the eigenvector The latter is the ith cointegrating vector , when , is non-zero. , and i = l,...,p), are thus solutions to the eigenvalue problem implied by the Maximuim Likelihood estimation of equation (7).

Table 5: Tests of the Cointegration Rank, 1970-91.

Table 5

Table 5 presents results from the test for the cointegration rank. There appears to be at least one, or maximum two cointegriting vectors. Note that adjustment for the degrees of freedom does not change the rank order. Visual examination of the cointegrating vectors (see Figure Al in the appendix) suggests that there is only one cointegrating vector and that it is stable even up to 1995. The last indication comes as a bit of a surprise given the high inflation and overall turbulence in the 1990s. Given that r = 1, the ; vector is


Where ; is an unrestricted cointegrating vector of parameters belonging to the variables denoted {p, e, m, w, p*}, respectively. As shown in vector , the parameters have been normalised with respect to consumer prices, since this simplifies the interpretation of the vector. It seems that long-run homogeneity from the theoretical model in section 3 does not hold.

It is necessary to investigate whether the cointegrating vector says anything about the structural economic relationships underlying the long-run model. This is done by imposing restrictions motivated by economic arguments (for instance, homogeneity restriction or variable exclusion) on the vector. Table 6 presents results from exclusion and stationarity tests on individual parameters, and for long-run homogeneity. The theoretical model in section 3 also implies that the parameters of the exchange rate and international prices are equal and this is also tested. Results show that these two parameters are equal and that it is necessary to include all variables in the long-run relation. This is consistent with the theoretical model. As previously shown, all variables are non-stationary. The tests also show that long-run homogeneity of the model in section 3 does not hold. Finally, Figure 2 shows the estimated error-correction term and supports the conclusion that it is stationary[13].

Table 6: Linear Restrictions on , 1970-91.

Table 6

Figure 2: The Estimated Long-Run Relation, 1.970-91.

Figure 2

5. Results

For the purpose of this paper and given the few observations, the inflation function is estimated with a single-equation approach. The model estimated is as follows:

Equation 8    (8)

where mt is the constant, Dt symbolises the dummy variables suggested earlier, (D1974, D1975, D1977), and et is a white noise error term.

Table 7: Over-parameterised Inflation Function, 1971-91.

Table 7

The general-to-specific approach is used to achieve parsimonious single- equation estimation for inflation[14]. Table 7 presents the results of the over- parameterised inflation function. At this stage the model is difficult to interpret in any meaningful way, its main function being allowance to identify the main dynamic patterns in the model and to ensure that the dynamics of the model have not been constrained by too short a tag length. Only the error-correction term, wages and money are significant.

Table 8: Single-Equation Estimation Explaining Inflation, 1971-91.

table 8

Table 8 presents the inflation function after variable reduction and linear transformation of the exchange rate[15]. The model is now a parsimonious characterisation of the data and all parameters have the expected signs. Schwartz-Information Criterion (SC) and the equation standard error (SE) have improved, while the number of variables has been reduced by half. At the same time, R2 has fallen oniv slightly. Estimation with the exchange rate in first difference has a positive contemporary exchange rate and a negative first lag with parameters of about the same size. Greater parsimony could be achieved by a linear transformation of these variables into the second difference of the exchange rate. The significance of the error-correction term indicates price stability over the estimation period. Thus, the condition of price stability for a successful financial liberalisation was fulfilled in Kenya. Only 11% of the previous year's divergence from equilibrium is corrected for in one year's time. This can be explained to a large extent by the existence of distortions in the Kenyan economy, including government controls of prices and wages, for most of the sample period. Regarding short-run dynamics, besides lagged inflation, wages have the largest impact on inflation. Thus, failing to take wages into account, as in Mwega (1990) and Ndung'u (1993), is bound to distort results.

The diagnostic tests are satisfactory, exclusion of the insignificant variables is accepted, and the forecast-test as well as the forecast Chow-test reject the predictive ability of the model for the period 1992-95. This is in line with previous conclusions, regarding periods of price stability and instability.

Figures 3 and 4 present the actual and fitted values, the corresponding scaled residuals, the forecast test, and the recursive estimations. The fit in Figure 3 is very good and the residuals are white noise, that is, until the period 1992-5 is included. The graph of forecast confirms the poor predictive ability of the model for the period after 1991. Results from the recursive estimation also show that parameters of the model were stable up to 1991. Parameter stability is also supported by the one-step residuals and one-step Chow-test.

Figure 3: Fitted Values, Scaled Residuals, and Forecast Ability, 1971-95.

Figure 3

Figure 4: Recursive Estimation, One-step Residuals, and Chow-test, 1971-91.

Figure 4

Table 9: Single-Equation Estimation Explaining Inflation, 1971-95.

Table 9

It only remains to show that by adding the years 1992-95 to the sample, the cointegrating relationship breaks down and parameters are no longer stable. Such a breakdown of the long-run relation is ascertained when the error-correction term becomes insignificant in the inflation model or when the adjustment coefficient approaches zero (Banerjee et al, 1993).

Results from estimating the inflation model over the full sample are presented in Table 9. The significant variables are lagged inflation, the money supply, the exchange rate, and the error-correction term. However, the significance of the latter is by itself not a sign that the long-run relation is stationary. That the adjustment coefficient is very close to zero indicates that the error-correction term is not stationary over the full sample. This breakdown of the cointegrating vector implies that agents switch to forward-looking behaviour, perhaps an indication of lack of credibility in the financial liberalisation process. This increases the variability of the price level.

It was also investigated whether the long-run relation breaks down when years are added to the sample period (1971-91), one at a time using the preferred model. The results are in Table 10. As shown by Table 10 the error-correction term is still significant in 1992, but turns insignificant in 1993 and 1994. In 1995 it is again significant, but with a much lower adjustment coefficient. It appears that there is still price stability in 1992, while in 1993 and 1994 this is no longer the case. As noted above, the adjustment coefficient, although significant, is practically zero for the full sample. Thus, adding the years 1992-95 to the sample implies a situation of price instability. However, this is something to be expected in an economy undergoing structural adjustment[16].

Table 10: The Development of the ECM, 1992-95.

Table 10

Figure 5 shows this absence of parameter stability and that actual inflation Is poorly explained by the model in the period 1992-95. It is also interesting to note that even when financial liberalisation is launched under conditions of price stability, the result can be skyrocketing interest rates and high inflation. This should not necessarily be interpreted that the financial liberalisation was the cause of the price instability. Reasonable explanations are the aid embargo that started in 1991, the sharp money growth around the election in 1992, and possibly the coffee boom of 1994.

Figure 5: Recursive Estimation, 1971-95.

Figure 5

6. Conclusions

The purpose of this paper has been to test whether consumer prices were stable enough to permit a successful financial liberalisation in Kenya. When undertaking financial liberalisation under conditions of price instability, interest rates might rise in order to offset anticipated inflation and to balance supply and demand for loanable funds. Risino, domestic interest rates may lead to large capital inflows and reduce borrower net worth. This may cause inflation and reduce investment and financial intermediation. Under such circumstances Implementing a financial liberalisation is difficult.

Price stability was said to exist when economic agents understood the process of inflation Generation and had some idea of how changes in the explanatory variables affected the consumer prices. It was pointed out that such a situation occurred when agents could rely on backward-looking, behaviour to make forecasts using a stable cointegrating vector. Cointegration between consumer prices, money supply, exchange rate, wages, and international prices was predicted by the theoretical model. However. a break down in the cointegrating relationship implied that consumer prices could no longer be accurately forecasted until a new stable equilibrium was reached.

Univariate examination of the full sample consumer-price series suggested a breakpoint in 1992, that is, after the interest-rate liberalisation. Since a statistically stable model could be produced for the sample period 1971-91. and cointegration was found for all variables, the conclusion is that the sample period, during, which financial liberalisation was undertaken, was characterised by price stability. This was also confirmed by recursive estimation of the inflation model. Thus, with respect to price stability it appears that the time was appropriate for embarking on financial-sector reform.

Since the cointegrating relationship broke down it can be inferred that economic agents could no longer forecast inflation with confidence using historical data. This breakdown implies that agents switched to forward-looking behaviour, perhaps an indication of lack of credibility in the financial liberalisation process. This, in turn, increased the variability of the price level. Even though Kenya's financial liberalisation was launched under conditions of price stability, the process nevertheless seemed to lead to high nominal interest rates and high inflation. Though inflation was subsequently lowered, high real interest rates persisted. These two occurrences should, however, not necessarily be attributed to financial reform per se, but to other causes both external and domestic. These included the destabilising aid embargo starting in 1991, the multiparty election of 1992, and, possibly, the coffee boom of 1994.


Figure Al: Long-Run Relations, 1971-91.

Appendix A, Fig 1


Banerjee, A., J. Dolado, J.W. Galbraith, and D.F. Hendry (1993), Co-integration, Error-Correction, and the Econometric Analysis of Non-Stationary Data, Oxford University Press, Oxford.

Barber, G.M. (1991), 'Consumer Price Indices and Inflation in Kenya, Technical Paper, No. 91-04, Ministry of Planning and National Development.

Bascom, W.O. (1994), The Economics of Financial Reforms in Developing Countries, St. Martin's Press, Inc., New York.

Bruno, M and J. Sachs (1985), Economics of Worldwide Stagflation, Basil Blackwell Ltd, Oxford.

Bruno, M. (1993), Crisis, Stabilisation, and Economic Reform: Therapy by Consensus, Oxford University Press, Oxford.

Fry, M.J. (1988), Money, Interest and Banking in Economic Development, Johns Hopkins University Press, Washington, D. C.

International Monetary Fund (various issues), IFS Statistics, IMF, Washington, D.C. Johansen, S. (1988), 'Statistical Analysis of Cointegrating Vectors', Journal of Economic Dynamics and Control, Vol. 12, pp. 231-254.

Johansen, S. (1991), 'Estimation and Hypothesis Testing of Cointegrating Vectors in Gaussian Vector Auto-Regressive Models', Econometrica, Vol. 59, pp. 1551- 1580.

Levin, J. (1994), 'Kenya: Two Steps Backwards and One Step Forward'. Macroeconomic Studies, No. 47/94, The Planning Secretariat, SIDA.

Lucas, R.E. (1976), 'Econometric Policy Evaluation: A Critique', Carnegie Rocliestet- conference Series on Public Policv, No. 1.

McKinnon, R.I. (1988), 'Financial Liberalisation in Retrospect: Interest Rate Policies in LDCs', in G. Ranis and T. P. Schultz (Eds.), The State of Development Economics: Progress and Perspectives, Basil Blackwell Ltd, Oxford.

McKinnon, R.I. (1991), The Order of Economic Liberalisation: Financial Control in the Transition to a Market Econoiitv, The Johns Hopkins University Press, London.

Mwega, F.M. (1990), 'An Econometric Study of Selected Monetary Policy Issues In Kenya', ODI Working Paper, No. 42, Overseas Development Institute, London.

Ndung'u, N. (1993), Dynamics of the Inflationary Process in Kenya, Doctoral Thesis, Gbtebor- University.

Republic of Kenya (various issues), Economic Survey, Central Bureau of Statistics, Nairobi.

Sharpley, J. and S. Lewis (1990), 'Kenya: The Manufacturing Sector to the Mid-1980s', in R.C. Riddell (Ed.), Manufacturing Africa: Performance and Perspectives of Seven Countries in Sub-Saharatz Africa, ODA, James Curry, London, Heinemann, Portsmouth.

Stiglitz, J.E. and A. Weiss (1981), 'Credit Rationing in Markets with Imperfect Information', American Economic Review, Vol. 7 1, pp. 393-4 1 0.

Swamy, G. (1994), 'Kenya: patchy, intermittent commitment', in 1. Husain and R. Faruqee (Eds.), Adjustment in Africa: Lessons from Country Case Studies, The World Bank, Washington, D.C.

Taylor, L. (1983), Structuralist Macroeconomics: Applicable Models for the Third World, Basic Books, New York.

Vandemoortele, J., and M.N. Gatangi, (1984), 'Kenya Data Compendium 1964-82', Occasional Paper, No. 44, Institute for Development Studies, University of Nairobi.

World Bank (1991), 'Pro-rarn Performance Audit Report: Kenya Industrial Sector Adjustment Credit', Report, No. 9746, The World Bank, Washin-ton, D.C.

World Bank (1992), 'Project Completion Report, Kenya: Financial Sector Adjustment Project', The World Bank. Washington, D.C.


[1]The author would like to thank Stephen A. O'Connell, Boo Sjöö, Qaizar Hussain, Steve Kavizzi-Mugerwa, and Dick Durevall for valuable comments and suggestions. The usual disclaimers apply.

[2]See Bascom (1994) and Fry (1988) regarding financial liberalisation in developing countries. With regard to Kenya's financial-sector reform, see the World Bank (1991, 1992).

[3]See, for instance, Taylor (1983) for a neo-structuralist view of financial liberalisation and its effects on the rest of the economy.

[4]This was the second attempt, the first one occurred between 1980-84, however, without much success (Swamy, 1994).

[5]The Kenyan government had already started reforming the financial sector, beginning in 1988. The reforms mainly concerned raising the interest-rate ceilings.

[6]When inflation is measured on a February-March year-to-year basis, which it is in official records, inflation is 29% for 1994 and 1.6% for 1995.

[7]All data. with the exception of the consumer-price series (CPI), are from various issues of the Economic Survey and the IFS. The CPI series are from Barber (1991), Republic of Kenya (1995), and Vandemoortele and Gatangi (1984).

[8]Recursive estimation means that the parameters are first estimated for a sub-sample. The parameters are re-estimated for the sub-sample plus one observation until the full sample is exhausted.

[9]More specifically, average wages are calculated as total earnings divided by number of wage employees.

[10]For this calculation I use the following countries: USA, UK, Holland, Italy, Sweden, Switzerland, France, Belgium, Germany, and the oil-producing countries. The weights are changed each and every year according to the country's relative importance in the trade basket (export and import). For all, but two years, at least 90% of the trade basket is covered.

[11]However, the first difference of the international price index is only significant at 10%. Since the estimated roots of the change in international prices are 0.37 and 0.22 for the two cases, respectively (D with and without a trend), both being far from unity, international prices are treated as an I(1) process.

[12]In fact, testing for order of integration by including 1994, but not 1995, still indicates price instability.

[13]The few observations available do not allow for a systems-based estimation approach. Thus, inflation will be estimated using a single-equation approach. Consequently, the short-run parameters and the adjustment coefficient will not be estimated with full efficiency. What is important for the purpose of this paper, however, is that the long-run parameters are obtained from a full information approach and this condition is fulfilled. Tests for weak exogeneity, that is restrictions on the vector, showed that only wages and international prices were weakly exogenous and that the other variables should be modelled.

[14]The definition of parsimony is to maximise the information from the model with a minimum number of variables.

[15]Note that although some variables (e.g. the international price index) were insignificant in the over-parameterised estimation, they turned out significant in the reduction process. This is most likely due to the presence of multicollinearity in the over-parameterised estimation.

[16]This is also what is implied by the so called Lucas-critique. Major policy shifts lead to radically different empirical estimates of econometric relationships (Lucas, 1976).

Previous Article Next Article

© 2001 Society for Policy Modeling. All rights reserved.
Published by Elsevier Science Inc.