Long-run forecasting seems difficult, especially in light of the challenges of short-term forecasting, such as whether we’ll have a recession, if inflation will come down, or if the Ukraine war will end. Fortunately, we don’t have to predict every twist and turn in order to predict the ultimate outcome. If I play a game of chess against a grand master, I confidently predict that I will lose. I cannot predict my opponent’s moves, but I’m sure I will lose.
Many business decisions depend on long-term conditions. Should an electric utility build a power plant with a large up-front cost and that will last 30 years or more? Should a bank go into another market, probably losing money for a few years but possibly finding good profits after that? Should a manufacturer invest in labor-saving equipment that will pay off only if wage rates continue to rise? Throwing up one’s hands and saying that projections are impossible doesn’t solve the business problem: should we lay out cash today in hopes of a future return? Most skeptics of forecasting have few alternatives to suggest. Simply assuming that future years will be like past years is itself a forecasting model.
Having advised on exactly the issues described above, I’ve found that three fundamental principles of forecasting can help any business analyst produce valuable long-term projections.
Model the Underlying Economic or Business Process
When looking at a long-term economic forecast that would drive an electricity demand projection, we can ignore short-run fluctuations. That allows us to ignore Keynesian or monetarist controversies and focus, instead, on the long-run drivers of economic growth: the labor force and productivity. Most of the people who will be working 20 years from now have already been born, so it’s not hard to estimate the size of the labor force. Productivity depends on characteristics of the workforce, such as education, as well as investment in capital equipment and technological change. The model is fairly simple and straightforward, though not perfect.
For local market projections, we can look at the drivers of population change. We have a pretty good understanding of what has led to migration in the past: job availability, housing costs, taxes and climate. We don’t have to get the details right to figure out whether one city will grow significantly faster than another.
The model we use to make long-term projections should be very simple, at least at first. Usually just a couple of drivers determine the long-run path. Ignore lesser factors.
Look at Historical Data
For long-run projections, we like to look at long-run data. Last year’s economic growth tells us little about the outlook for the economy 10 years from now. But looking at very long growth trends does give us a sensible range of what we might expect. With a little digging, long-run data can be found on many issues relevant to business decisions.
When looking at long-run data, the first question to consider is if the trend has shifted in the past. If an industry’s sales have averaged five percent growth in some decades, but ten percent growth in other decades, that is important information. Trying to understand those shifts will be important.
The next element to examine in long-run data is how wild are the short-run fluctuations. Sometimes the year-to-year figures are very close to a trend line, but in other cases they vary widely.
Then it’s time to iterate. Does our look at long-run data give us more questions about what drives the underly process? If so, it’s time to create a somewhat more detailed model of the underlying process. But if the simple model seems to capture the trend, then it’s good enough.
Forecast With Humility
No business or economic forecast will be perfect. Recognizing that reality will lead to better business decisions. Television pundits may pound the table and profess certainty. That makes for good theater but poor decisions. The world is incredibly complicated, and nobody will get everything right time after time.
The business decision that is based on the forecast should be made recognizing that no forecast is perfect. The anticipated profit must be great enough to cover the risk that the projection driving the decision may be off.
The humble approach looks at the forecasting track record as it evolves. Errors of the forecast are considered in two ways. First, is this forecast error of a size that we would expect? Here it’s best to look at the absolute value of the error, not the direction. Graphing the historical data with a trend line and the forecast helps.
The second question to ask about our forecast errors regards bias: Are we persistently over-forecasting or are we under-forecasting? Finding that our errors are all in the same direction should lead us to alter the underlying model or the parameters of the model. Some decision-makers want conservative bias in the forecast, so that sales projections are somewhat low-balled. Better practice is for the forecast to be the best possible, then conservatism inserted into the decision-making process. That helps the forecasting analyst make better projections over time.
Long-run projections are necessary because most business decisions depend on conditions that are more than a year or two away. The better the projections, the better the decisions will be, and the better the company’s long-run profits.
Read the full article here
