My previous blog ended with a question about the usefulness of anticipating the long-term future if that future is highly uncertain. Ever since the 1982 article on “Trends and random walks in macroeconomic time series” by Nelson and Plosser, there has been a debate about the long-term statistical properties of GDP and other macroeconomic variables. Nelson and Plosser could not reject the hypothesis of a random walk (with drift), which means that random shocks have a permanent impact on the level of GDP and that the uncertainty interval around forecasts becomes wider and wider with every year you try to peek farther into the future. The message seems to be: If next year’s world is already very uncertain, don’t even bother forecasting the world in 2030.
Others found that “macroeconomic time series are best construed as stationary fluctuations around a deterministic trend function”, if you allow for a few structural breaks in the trend. The consequences for long-term forecasting are huge because, in this case, random shocks are transitory, there is mean reversion, and it is in fact easier to analyze long-term trends than short-term fluctuations.
So, when we look 20 years into the future, do we buy into the predictability of long-term trends? Actually we do not. I am ready to assume that the long-term future is virtually unpredictable. Even if you are convinced by the finding of deterministic trends, it will be tough to predict the structural breaks of the future. So, we don’t even pretend that we can predict the next twenty years with some kind of accuracy. Every time we are confronted with this uncertainty in our long-term, forward-looking analysis, we preempt any further criticism by emphasizing that these are scenarios and should not been read as forecasts. That seems to be a cheap shot. We seem to think that we can tame the beast by just changing its name. We put a lot of effort in discussing a future world, but we appear to avoid responsibility for the realism of the exercise. It seems not only cheap but also convenient because, as Kaushik Basu put it in his Foreword to the Global Development Horizons report, “one advantage [the report’s analysts and researchers] have is that, by the time the accuracy of their forecasts becomes known, most of them will have moved on to other pursuits”.
So, why do we go through all the trouble of exploring emerging trends and creating future worlds with complicated models? The reason is simple. With all the uncertainty about the future, one thing is sure: the future will be very different from the present. If policy makers get stuck in the present, let alone being stuck in the past, they will certainly not be able to accommodate future trends. The work on China 2030 that I was involved in, was a good example. If China’s policy makers would focus on the present, they would not feel the urge to change their policies because, by many different measures, China is pretty successful. However, because they anticipate (and aspire) a very different future with new challenges, they are willing to prepare for radical changes in policies. And this is true all over the world. From health care reform to trade policy, you don’t get it right if you merely analyze the current situation. With all the uncertainty, the future reality is better understood by exploring future trends than by studying the present.
When you explore future trends, you must do it in a systematic way. And that is where scenarios and models come in. Scenarios are not set up as forecasts; they are thought experiments. There are basically three different approaches. One is the bottleneck approach. Extrapolating trends will reveal bottlenecks (scarcity, congestion, pollution). That extrapolation might not be realistic because the bottlenecks trigger, in many cases, endogenous solutions, but as a warning sign they can be very helpful. An extreme example of trend extrapolation was provided in 1894 (when horse cars were used for transportation) by a writer for the Times of London who predicted that in 50 years every street in London would be buried under 9 feet of manure. The problem with that example was that it completely ignored feedback mechanisms, but it still made a point. In our extrapolations, we use general equilibrium models to take into account some of the key endogenous feedback mechanisms. A second approach is the planning approach. Define where you want to be (controlling climate change, becoming a high income country) and think back what might be needed to get there. Again, you need models to help you with this exercise. Again, it might not generate an accurate forecast, but such a scenario is a useful tool for policy makers. The third approach was probably first developed by Shell. It is like a war game or a fire drill. Build a scenario in which you expect the unexpected and think the unthinkable. Then ask policy makers to think through their policy reaction. Even if that scenario never materializes, it is useful to prepare policy makers for a changing world and to “shake their mental map”. Again, you need structural models to simulate policy responses.
It is in this tradition that we do long-term analysis, with elements of all three approaches. We use scenarios, not to cover up our inability to predict the future, but to help policy makers escape the shackles of the past.