A common misconception among market analysts is that the South African Reserve Bank (SARB)’s QPM is a forecasting model. QPM is a storytelling device. It is used to provide an assessment of the state of the economy and its outlook given a range of assumptions and judgements and provides guidance on the appropriate path for policy to achieve the Monetary Policy Committee (MPC)’s objectives.
A recent SARB working paper provides a welcome update of its QPM model used for monetary policy decision-making.
In this post, I discuss the implications of the changes that have been made and how the new version of QPM helps to improve our understanding of the determination of macroeconomic outcomes and the appropriate monetary policy stance. It remains to be seen how the new version of the model will be used for communicating the policy stance. I argue that SARB should be clearer about how QPM is used to encode the economic narrative of the MPC when thinking about the appropriate stance of policy.
In reporting on the model update, many market analysts have focused on whether the model will be more or less ‘dovish/hawkish’ – i.e. whether the model will require less/more aggressive changes in rate setting to achieve the inflation target – and whether its forecasts have been more accurate. This represents a misunderstanding of the role of QPM in the forecasting and decision-making process. The stance of monetary policy is determined by the MPC, not by QPM. Different policy settings could be consistent with the model structure and depend on the trends and ‘unobservables’ imposed on the model. Economic unobservables describe the Bank’s assessment of the cyclical position of the economy and its assessment of the stance of monetary policy. The model, therefore, is a tool for communicating the economic narrative of the forecasters (or MPC). It is not an independent forecasting model.
The forecast performance of the updated model is also driven by the various assumptions and judgements imposed on the model. One should therefore not assess its forecast accuracy without consideration of the staff and MPC assumptions that affect the assessment of the state of the business cycle implicit in QPM projections, the shocks that have recently hit the economy and the judgements imposed on the paths of various inputs into the model.
The paper focuses on the adjustments to the structure of the QPM and does not elaborate on the evolution of the SARB’s views of the unobservables in the model. Given the recent misalignment between the MPC’s published projections and actual MPC settings, the model update provides an opportunity for SARB to re-align the model with actual MPC views to make it a more reliable guide to future MPC decisions. Unfortunately, little detail has been made available about how the SARB has updated its views about the trends and equilibria that affect its forecasts and policy assessment. One hopes that SARB will provide such details when it makes monetary policy decisions in future.
Nevertheless, the update as described in the paper is important for several reasons:
- the relationships between economic variables have changed since the model was first implemented in 2017
- the original model did not include several important channels that affect the South African economy (such as the impact of fiscal policy)
- the determination of several key ‘unobservables’ that matter for assessment of the appropriate stance of policy needed refinement (e.g. updating the factors that determine the neutral interest rate estimate)
- it is an opportunity to better align the model with the preferences of policymakers in achieving policy goals, as in the calibration of the Taylor rule that determines how policy should be adjusted to meet the inflation target.
In what follows I discuss some of the key changes to the model.
- Inclusion of a Fiscal Block
The updated model includes a fiscal block, which makes it possible to characterise the link between South Africa’s fiscal policy, business cycle, and the risk premium. The paper includes a summary of the transmission of fiscal policy to the economy. The benefit of the inclusion of a fiscal block will be better storytelling ability (such as providing richer explanations for wage or inflation pressures) and the ability to characterise the implications for monetary policy. The model’s interpretation of the stance of monetary policy is that fiscal policy has been loose in the wake of the global financial crisis (GFC), consistent with our assessment and that of the IMF.
The paper does not say much about how the fiscal block will be used in practice, such as whether the model will take all of the Treasury’s public finance projections as given or use specific projections from the SARB instead. What SARB will do in practice might make a difference to SARB’s overall projection given the large size of historical public finance forecast errors. The National Treasury has, for example, consistently underestimated the deterioration in our public debt. So it will be interesting to see how SARB incorporates Treasury projections and adjusts its view of the target debt level over time.
It would also help a lot if SARB could begin publishing quarterly GDP projections (real and nominal), given how difficult it can be to forecast the GDP deflator. The budget balance data used is for the main budget, so when forecasts begin being presented it will likely raise questions over the implications of the treatment of Eskom and state-owned enterprises by Treasury in the budget (given that the recent IMF Article IV treated Eskom support as a capital transfer above the line (i.e. as an expenditure item), while Treasury treated it ‘below the line’ in the February budget i.e. they did not include such spending in the main budget).
- Risk Premia
The risk premium specification in the QPM now incorporates not only public debt but also expected risk compensation required by international investors (proxied using the difference between the expected stock market variance for the US S&P500, the ‘VIX’, and its assumed steady state level). This is an improvement over the previous version of the model, but more evidence on how well the new specification aligns with observable measures of South African sovereign credit risk and the historical consistency between the modelled risk premium, interest rates and public debt trajectories would have been insightful.
The term premium specification used in the updated QPM assumes the same drivers as for the risk premium. That is, QPM now assumes that fiscal policy is a key driver of the term premium. That might be considered a little controversial since measures of South Africa’s country risk do not explain the dynamics of the term premium.
The SARB’s observed measure of South Africa’s risk premium –
the JP Morgan Emerging Markets Bond Index Plus spread – is also contaminated by co-movement with global factors and a measure purged of such influences would likely better approximate the country risk premium concept used in QPM. Such a measure would also have a higher correlation with the South African term premium.
- Neutral interest rate
The neutral real interest rate is a useful concept for assessing the stance of monetary policy. SARB models neutral as a function of an estimate of the global neutral rate, its South Africa risk premium estimate and the expected change in the real effective exchange rate. The neutral rate is determined now as the outcome from satellite models that determine the these components, without forecaster intervention as was the case before.
As an unobservable, economists will always disagree about the value of the neutral rate. The update is an improvement as market analysts can now better understand the drivers of SARB’s estimates. Interestingly, the inclusion of the fiscal block has not meaningfully changed SARB’s updated neutral assumption. With that in mind, our estimates suggest that market pricing has long implied that South Africa’s neutral rate might be higher than SARB and many private sector economists have been assuming. Our updated estimate is currently at 7.5% nominal – implying 3% real (compared to SARB’s assumption of around 2.5%).
- Policy Rule
The Taylor rule in the model determines how policy should be adjusted to meet the inflation target. The rule was updated to focus on inflation at a forecast horizon at 9 to 12 months into the future, and now includes the growth gap between their one one-year-ahead forecast of GDP growth and potential. Together with an updated calibration to better describe historical MPC decisions, the updated rule implies slightly stronger policy reactions to deviations of inflation from the target. It would have been useful to present a decomposition of the policy advice from the Taylor rule into its components to help demonstrate the relative contribution from the smoothing parameter compared to the other drivers in the rule, as well as how well the updated rule describes historical policy decisions. A separate working paper or occasional bulletin article on this issue would be very useful.
- Wage and inflation determination
The updated model includes a new specification for unit labour costs, now distinguishing between private and government; the core inflation Phillips curve now includes the impact of global oil prices as well as fuel and electricity spillovers in the services and core Phillips curves; and updated inflation expectations determination with a higher weight on the inflation target.
The changes to unit labour cost in the model are important because there are big differences between developments in the public and private labour markets: aggregate public sector wages grew much faster than private wages over recent years and there have been persistent differences between unit labour costs and wages.
However, there is no specific analysis of how the Phillips curve specification changes themselves affect the model’s ability to explain inflation dynamics. In earlier work, we showed that the deterioration in Phillips curve fit was affected by the output gap measure used by MPC, so it would be useful to understand whether the recalibration and updated output gap estimates have improved the fit of the model specifications.
- Credit channel
The SARB has also added a measure of bank lending conditions to QPM’s cost of credit channel that affects aggregate demand. This brings the model a bit more in line with best practice, recognising the importance of time-varying lending spreads in affecting policy transmission. That said, it might have been better to separately account for bank funding spreads and lending spreads, instead of proxying these as the difference between bank lending rates and the prime rate. This matters in the context of heightened sovereign risk, as bank funding costs are a channel through which fiscal risk can transmit to the economy and affect pass-through of interest rate changes to loan and deposit rates.
- More on Model Calibration
Changes to the structure of the economy and the estimated relationships between economic variables since the model was first implemented in 2017 required the model to be re-calibrated to better fit South African data. To assess the model calibration, one can assess:
- whether the model’s impulse response functions conform with our understanding of economic relationships
- the forecast accuracy of different specifications and the drivers of forecast errors using the model structure
- the consistency of different model-implied trends (i.e. whether the equilibrium exchange rate is consistent with the trend in imported inflation, for example)
- shock decompositions to understand how different economic shocks have contributed to deviations from model equilibria and whether the economic narrative is credible
Unfortunately, the South African economic literature provides little guidance on how key parameters should be calibrated to be consistent with the mean values in the data. Economic trends have also shifted a lot since the model was introduced. For example, trend growth of potential output is assumed to be 2.5 percent, and it is difficult to know whether this is realistic, given the trend decline in economic growth since 2010 and uncertain outlook for resolution of our electricity constraints. More academic research into the structure of the economy would be helpful for informing the appropriateness of the model calibrations.
Overall, no detail is provided on how different series are filtered (i.e. how trends in the model are determined), which can be important since different filtering approaches can make a material difference to the trends obtained. The paper also makes no mention of the role of nowcasts from satellite models in the forecasting process either.
The paper provides forecast accuracy comparisons between the versions of QPM and impulse responses for key concepts in the model to show that the update enhances forecasting ability a little. Unfortunately, the paper does so without stripping out the role that forecaster assumptions and judgements about the outlook for the economy might have played in that assessment. It is also not clear how much the unobservables in the model have been updated in the revision. The exchange rate equilibrium, risk premium and potential growth estimates are from satellite models, so these and the neutral rate are not estimated directly using the structure of QPM.
That brings us to the importance of the economic narrative implied by QPM forecasts.
- The Economic Narrative
Judging whether the new version of the model is an improvement requires assessment of whether it provides a credible economic narrative about historical developments in the economy. The paper includes a section that summarises how the model interprets the impacts of different economic shocks relative to the previous version of the model, which helps one understand the key relationships the model captures.
Under a re-calibration of QPM with equilibria filtered out to be consistent with the model structure (i.e. letting QPM determine unobservables instead of imposing them exogenously), the estimated neutral rate or other key unobservables would be different to what has been presented. It would be interesting to know how different these concepts might be. It would also help to provide an assessment of whether their forecasts are based on economically consistent gaps: e.g. whether their exchange rate gap is consistent with the output gap and the projected policy path, by way of example.
The model update has provided no evaluation of the SARB’s historical business cycle narrative. Best practice among inflation targeting central banks is to periodically assess what aspects of the central bank’s assessment of the drivers of the economy have been spot on, and which aspects the Bank got wrong. This helps improve policy setting and forecasting by helping the Bank learn from its mistakes and better characterise the changing relationships among economic variables.
By way of example, SARB has gradually revised their judgements about the underlying trends in the economy. SARB’s terminal projections of potential growth rate has declined over time, while its terminal neutral interest rate projection has risen. One hopes the SARB will explain these revisions through the lens of QPM in future publications.
It is also a pity that there is no structural analysis of the economic drivers of forecast errors. For example, it would have been helpful to show the contribution of demand-, supply- and monetary policy shocks to the SARB’s inflation and GDP forecasts. This would help one assess the appropriateness and impact of the SARB’s front-loaded policy easing at the onset of the COVID-pandemic, and the recent front-loaded tightening.
Such additional information would help analysts evaluate whether the narrative underlying the model forecasts are reasonable. Without such detail, it is hard to assess what the updated model implies for drivers of forecasts compared to the previous version and therefore the appropriate stance of policy. Hopefully, the release of the updated model will open the door for more QPM-related research outputs that help the market understand the SARB’s policy assessments.
