MEFMI Develops a VAR Model for the National Bank of Rwanda

The National Bank of Rwanda (BNR) is planning to adopt a forward-looking price-based monetary policy regime.  This requires the scaling up of macroeconomic modelling and forecasting as well as economic analysis skills. In light of this requirement, the Bank requested MEFMI to conduct an In-country capacity building workshop for its officials.

The main objective of the workshop was to reinforce macroeconomic modelling and forecasting skills to the BNR staff. The workshop was facilitated by Mr. Sayed Timuno, MEFMI Programme Manager and the Mr. DeJager of the South African Reserve Bank in two stages. The first stage was conducted from 18 to 22 June 2018, in Kigali, Rwanda with the second stage scheduled for later this year.

Officials from the Bank expressed gratitude to MEFMI for developing a Vector AutoRegression (VAR) model for the Bank. The workshop also developed a Vector Error Correction Model (VECM) template for use by the Bank.

The workshop was attended by 16 participants from various departments of the Bank. 19 percent of the participants were females. The gender distribution is indicative of the staff composition in the Bank, specifically in the Macroeconomic Modelling and Forecasting Units. It should be noted that this composition is not unique to the BNR. The Programme has witnessed that, in the MEFMI region, participation in activities related to macroeconomic modelling and forecasting is mostly skewed towards males.

The training equipped participants with requisite knowledge and analysis of how macroeconomic modelling tools can be used for macroeconomic management and policy decisions. It was delivered through a combination of lecture presentations, demonstrations and discussions. The ratio of 30:70 between lectures and practical sessions complemented by individual hands on exercises was effective for delivering a technical course of this magnitude.

Participants were taken through different economic theories and shown how they related to macroeconomic modelling and forecasting. The identification of macroeconomic variables used in the estimation process was also covered. Hands-on exercises focused on loading data, analysing and conducting descriptive statistics into the Eviews with the guidance of the resource team.  Participants were also taken through properties of least square estimators, and different statistical inferences. The team also discussed how the sample size can impact the macroeconomic model. Each participant was then required to do individual exercises on single equation estimation.

The two Step Engel Granger equation estimation and forecasting involved testing the stationarity of variables. This was followed by assessing cointegration by creating residuals and then checking if they were also stationary. An error correction model was then developed and tested for forecast accuracy against the single equation model which was estimated earlier. The comparison was done using the root mean squared error. Model diagnostics and forecasts were also conducted

Discussions were held during the diagnostic session, focusing on estimation diagnostics such as residual, stability and coefficient tests. The residual test covered heteroscedasticity, normality and serial correlations. Coefficient testing focused on linearity test (i.e. adopting the wrong functional form), testing for the omission of relevant variables and testing for inclusion of redundant variables. Stability tests focused on Ramsey RESET and the CUSUM tests.

The concept of VARs, process of estimating global VARs, reduced form VARs, and structural VARs were demonstrated to participants. In addition, the resource team also took participants through granger Causality Tests, Johansen Cointegration Test, Impulse Responses and the variance decomposition. Hands on exercises focused on the estimation of VECM. Participants were also taken through the estimation of single unit BVAR models. They were shown how to obtain posterior estimates of the coefficients of the model for each unit and how to obtain estimates of the mean effect across units. The resource team also described a procedure to endogenously group units with similar characteristics as this was useful, particularly when one wants to distinguish the impact of certain shocks on selected variables, or when policy advice requires some particular endogenous classifications. The resource team took participants through VECM estimation. The starting point was testing for stationarity. Variables were found to be stationary at levels, suggesting that the use of ordinary least squares was not appropriate for estimation. Participants were therefore introduced to the Johansen’s technique. Cointegration tests, lag length structure, impulse responses and short run dynamics were conducted. Participant also prepared forecasts with this model.

Various ways of reporting forecasts from the estimated model were demonstrated to participants. They were also required to do point forecasts, density forecasts and fan charts in Eviews. The resource team assisted participants to develop a VAR model for Rwanda. This was done using Rwanda Data. The model was tested and performed well. Impulse responses were conducted. The model was used for forecasting and it produced a good results for 8 quarters. The model is based on three (3) variables; namely Money Supply, Real GDP and CPI. It looks at responses of GDP and CPI due to a change in Money supply. The development of this VAR model will go a long way in assisting the Bank to conduct various scenarios and offer evidence based advice to BNR Management

The workshop was officially closed by Mr. Timuno. In his closing remarks, he highlighted that MEFMI was committed to capacitating its client institutions. He encouraged all participants to discuss pertinent issues relating to structural breaks in the Rwanda time series data.  He also encouraged them to continue practising what they had been taught and to interact with MEFMI to further improve their knowledge in this area. 

One recurring recommendation from the participants as captured in the evaluation of the workshop is the need to extend the course to two (2) weeks.