**Table 1.**

utilizing the Bounds Test, does not require this pre-testing of the order of integration. It uses the F- and t-statistics to test the significance of the lagged variables in a univariate error correction system without establishing the order of integration of the data generation process underlying the series. However, it is important to establish the order of integration beforehand to ensure the absence of I(2) series since their presence would violate the properties of an ARDL model which requires variables to be only I(0), I(1) or they should be mutually integrated. The ARDL methodology also has an advantage over other methodologies in that the its parameters can be estimated consistently without invoking exogeneity and residual serial correlation, especially if the order of the ARDL is appropriately augmented by the

This being the case the objective of the chapter could therefore be investigated by using the ARDL model and Error Correction Model (ECM) frameworks. In this regard Eq. (1) can mathematically be specified as an ARDL model with *p* lags of *inf* and *q* lags of *X* (where *X* is a *kx*1 vector of independent variables, which include in this case money supply growth, lending rate, nominal exchange rate, import prices,

*q*

*i*¼0 *α*0 *i*

*q*

*i*¼0 *α*0 *i*

where *t* is the time period, *θ<sup>i</sup>* are *kx*1 coefficient vectors; *α<sup>i</sup>* are scalars and *ε<sup>t</sup>* is a

*<sup>θ</sup><sup>i</sup> <sup>Δ</sup>inft*�*<sup>i</sup>* � � þ<sup>X</sup>

*X<sup>t</sup>*�*<sup>i</sup>* þ *ε<sup>t</sup>* (2)

*ΔX<sup>t</sup>*�*<sup>i</sup>* þ *εt*, (3)

suitable specification of the lag structure of the variables [25].

*Linear and Non-Linear Financial Econometrics - Theory and Practice*

fiscal deficits and output gap), ARDL ð Þ *p*, *q* as:

*inft* <sup>¼</sup> <sup>∅</sup>*inft*�<sup>1</sup> <sup>þ</sup> *<sup>γ</sup>*<sup>0</sup>

**4. Model estimation and results**

**3.2 Data**

**258**

*inft* <sup>¼</sup> <sup>X</sup> *p*

*i*¼1

parameterized and expressed in error correction model form as:

*Xt* X *p*

*<sup>θ</sup><sup>i</sup> inft*�*<sup>i</sup>* � � þ<sup>X</sup>

disturbance term with a zero mean and constant variance. Eq. (2) can be re-

*i*¼1

where ∅ is the speed of adjustment and *Δ* is the difference operator.

The paper uses quarterly time series data for analysis for the period January 2001-June 2019. The data for all the variables is obtained from Malawi's Central Bank, the Reserve Bank of Malawi, except import price index which is obtained for the Economist Intelligence Unit database [24]. Real GDP were in annual frequency and had to be interpolated to transform them into quarterly data frequency.

The estimation of the model starts with the examination of the time series properties of the data, using the Augmented Dickey-Fuller test to test for stationarity of the variables. The results of the unit root tests indicate that all the variables, except output gap and money supply growth are integrated of order 1 (I(1)) at below 5% significance level (**Table 1**). This means that in this model some variables have integration of zero order, I(0), and others have integration of the first order, I(1). This result satisfies the requirement for the application of the ARDL methodology which is further supported by the results of the Bounds test (**Table 2**). The result of the Bounds test for co-integration proposed by Pesaran et al. [25] within the ARDL framework, shows that the null hypothesis of no

*The results of the augmented Dickey-Fuller unit root test.*


#### **Table 2.**

*Bounds test results for co-integration relationship.*

equilibrium level relationship is rejected at below 1% error level by the F-test statistic (**Table 2**).

The Bounds test results in **Table 2** show that F-statistic has the computed value of 8.96 which exceeds the upper bound value of, I(1) which is 3.99 at 1% level of significance, implying that inflation rate and its determinants in the model are cointegrated and approach the long-run equilibrium, calling for the application of the ARDL approach [25]. The implication of this is that the parameters of the model can be estimated consistently without invoking exogeneity and residual serial correlation. The parameter stability test based on the plot of the cumulative sum of recursive residuals squares (CUSUM test) and the plot of the cumulative sum of squares of recursive residuals show that the estimated parameters of the ARDL specification are stable at least over the study period (**Figures 1A(a)** and **(b)** in the Appendix).
