They are just a part of policy decisions — they are not the policy
In her most recent Business Day column Mamokete Lijane argued that the economic models applied by policymakers are calibrated on the prepandemic structure of the economy, which no longer exists and will therefore lead to wrong forecasts and policy responses, and undesirable economic outcomes (“Policymaking based on the past can lead to big, persistent errors”, June 22). In defence of economic models and their use, I wish to disprove this interpretation.
Lijane concluded her piece in no uncertain terms by shooting down the SA Reserve Bank’s quarterly projection model (QPM) as outdated, and suggested its policy rate guidance should be ignored. I argue that interpretation of economic models in general, and their use, lack the nuance that is involved in economic modelling, and that her conclusion that model outcomes should just be ignored is simply wrong.
Let me outline the areas where I agree with Lijane. First, I have no contention with the statement that economic shocks such as the global financial crisis and Covid-19 affect economies in structural ways that change economic relationships and that existing economic models cannot capture. In fact, take that point of view to the extreme and aspects of the economy are never captured by economic models.
Second, I concur that model outcomes that poorly predict the evolution of economic variables can lead to wrong policy responses, especially in cases where there is an overreliance on them. That may well have been the case over the past decade in SA and globally. Economic growth, and therefore tax revenue collections, were overestimated, while expenditure overruns were a permanent feature, leading to persistently higher deficits and debt-to-GDP ratios.
My pushback to Lijane’s main conclusions aims to provide the nuances in economic modelling. Models are a part of the problem she highlighted as far as wrong forecasts are concerned. The other part is the institutional set-up, which is problematic. I remember a discussion about economic growth forecasts almost a decade ago when a senior nontechnical leader rejected a forecast because it was too low, with no rational basis.
There were no disagreements in terms of the model assumptions that ultimately drove the output, but the output that was sensibly produced by technical, economic modellers was not accepted. This demonstrates that in a poorly calibrated institutional set-up the advice of technical people can be ignored for political reasons, leading to suboptimal decisions. This is not a models issue but one of institutional set-up.
By design, economic models almost never capture the full functioning of the economy; they have reduced representations of how the economy work. Their usefulness lies not in accurately predicting economic variables to the second decimal point, but in first getting the direction of the evolution of economic outcomes right, and the order of magnitude of the outcomes. For example, the consequences of differences between a forecast of 4% and 4.2% economic growth are inconsequential for economic policy.
Though it is true that the structure of the economy is changing in ways models cannot capture, it is not true that models have remained static. National accounts data is rebased, and models are re-estimated to capture the changing economic structure, still in the spirit that they are a reduced representation of the true functioning of the economy.
The more important nuance Lijane misses is one Chris Loewald, the chief economist and a monetary policy committee member of the Reserve Bank, made in this newspaper in October, which is that models are just part of policy decisions — they are not the policy (“Why quarterly projection model is just part of monetary policy moves”, October 14, 2020). There is a lot of judgment that goes into policymaking, way after the economic forecasts have been produced and approved.
In a world with so many moving parts, models provide the framework for thinking about the economy. And no, their forecasts must never be ignored but must be overlaid with judgment from information that cannot be captured in the model.