In 2010, Carmen Reinhart and Kenneth Rogoff, two respected Harvard economists, shot to fame for publishing a paper that seemed to support the view that once a national debt had crossed the 90% GDP threshold it would impact negatively on economic growth. By 2013 a coding error in the data spreadsheet had been discovered disproving this conclusion. There was no such relationship.
At a more mundane level, Cisco’s projection of Internet traffic for the next five years is impressive. Too impressive for most other forecasts, which it exceeds by several orders of magnitude. Similarly with McAfee’s estimates of economic losses from IPR theft. Do the politicians and business executives care too much so long as the figures they cite support their case? It depends whether investment decisions hinge upon them. They did, for example, during the dot.com bubble.
Taken seriously, forecasting is an econometric skill. It involves various stages, such as developing a model, testing the model, revising the model in terms of the categories of data actually available (which is a major problem because the nature of the data actually available may not match the theoretical category being captured in the model), collecting the data from often unreliable or unclear data sources, massaging the data to iron out irregularities (which immediately contaminates the data), inputting the data (without messing up), and then running the equations. Every stage is fraught with problems. Building the model is itself as much an art as a science. Guessing what time lags to experiment with, what elasticities to play with, which factors may or may not be independent of others, etc., is the science of the imagination. Build in behavioural assumptions and the model will ignore others. In other words it is biased or tilted from the outset. Change the specifications to accommodate the types of data available and the model loses its original perspective. The results can never outdo the ingredients.
Astrophysicists probably come closest to real forecasters because they rely so heavily upon mathematical logic and the observations of others. They have the luxury of being successful when they are wrong, because that’s how science progresses. It never works like that in economics because you can never be proven wrong. There are always explanations for why you are right, and other factors (ceteris paribus) that intervened beyond the control of the model. Economics can never be more a science than an ideology, a strong belief in the veracity or misjudgement or manipulation of markets. For example, the interconnectedness of economic activities suggests that the whole is greater than the sum of its parts, that one person’s expenditure is another’s income. On the contrary, the Chicago School and its acolytes would argue that the expectations of individuals over-ride the impact of social actions. Where are we today? The Keynesians say there is a liquidity trap, meaning that even though there is enough cheap money around no one dares invest it or spend it because no one else is doing so. The state must fill the gap until confidence is restored. The anti-Keynesians say that no one will believe that state spending will help, it will just add to the national debt and to inflationary expectations, because inflation reduces the value of debt. Everyone, well almost everyone, agrees that austerity produces unemployment and piles on the misery, but some see this as a self-inflicted wound while others as catharsis.
What is true in all markets is that money can buy the best advice available. Well, unless you are unfortunate enough to encounter an economist of the other persuasion. In that case go for the cheap version of forecasting: extrapolation by consulting companies. You pay less and no one remembers the forecast when it proves wrong in 5 year’s time. And they are always around 5 years; short enough to impress the BoD, long enough for everyone to have moved on. Especially suitable for the ICT industry. How much more interesting therefore if models were used to predict the past, using the best available data. Enter it into the equation, press GO and history is foretold! My model says that those forecasts in the 1940s that by the end of the century there would be a computer in every city were not wrong.