Introduction to Forecasting
INTRODUCTION
This essay introduces some considerations about using forecasting as a tool of corporate planning. It discusses some of the issues. It is not a tutorial to make you an econometrician but to give you some background to evaluate forecasts.
How economic variables are forecasted depends on the time-period forecasted, what is being forecasted, and why the forecast is made. It also depends on the past and present data available.
It is important to remember that any forecast is uncertain. There is a range of uncertainty that maybe can be estimated, or maybe not. Keep this in mind when reading about any forecasted number.
Data. Forecasts and decisions are based on data. Just because the data are numbers does not mean they are accurate. There may be defects in the past data. There may be definitional problems, data collection problems, data summary problems, and data analysis problems.
One set of reliable data is internal corporate data. But not necessarily external data. Individuals lie on surveys and about their income. Corporations have many ways to legally lie about their sales, profits, financial position, and everything else. Countries set economic targets and then lie about the economic statistics they report. Revisions of macro data can be large, even in the short run. There are also long run revisions, sometimes as late as 10 years later. Economic data from many countries are only approximate.
Time Period. Corporations typically make short-range forecasts such as monthly, quarterly and yearly sales as part of corporate planning. Actuals are compared to budget figures. Reasons for differences are discussed.
Companies have reliable data but be careful of the accounting rules that created them. For example, land and buildings are carried at original cost; buildings are depreciated even though their market value (opportunity cost) may be greater. Profit measurements are distorted, which may affect investment decisions.
There are some simple ways to forecast. Assume a company produces consumer goods in a relatively stable market, at least over the next year. The first way is to use income elasticity of demand. From past data, compute how the yearly unit sales of a product or product group increased with the increase in total real income. Say, for example, that on average for every 1% increase in real disposable income, unit sales went up 2%. So, if you (or the average of other forecasters) expect a 3% increase in real disposable income next year, your forecast is that the unit sales of the product will increase 6%. You might adjust it if you believe there are changes in other factors that influence demand. You might also use this as a benchmark; if you increase advertising but sales still go up 6%, you might doubt your advertising is effective (a waste of money).
Another approach is trend, or time-series, analysis. Only look at past sales; ignore factors influencing sales. Compute the yearly trend over time; there are statistical programs to analyze the trend. This is often a good starting point. Simple estimation packages are available.
You may have a yearly forecast but you know from past experience there is seasonality. Use it to break down or apportion the yearly trend by month or quarter. The U.S. government uses seasonal adjustments to much of its data; adjustment packages are available.
Much more data is now available, including 80 years of post-World War II data. Used carefully, longer time-series data usually gives you more confidence that shorter time-series data.
Sophisticated programs analyzing sales data captures much more information on individual customers and their buying behavior. A company like Amazon tracks individual buying behavior and “suggests” repeat purchases and purchases of similar or related products and services. Companies like airlines can also use this data to practice dynamic price discrimination, changing prices and the mix of seat categories on individual flights.
Don’t forget Pareto’s Law. Use internal and market data to discover the 20% of your customers or product line that account for 80% of your sales or operating profit. You may also discover some of your customers or products are unprofitable. Raise prices or stop supplying them. Use company resources elsewhere.
Long-term forecasts. For all forecasts, the forecasted number has a band of uncertainty – a confidence interval or a statistical distribution (often assumed to be a normal distribution or bell-shaped) around it. You should assume that the more into the future a forecast is, the wider the confidence internal (a measure of uncertainty).
Forecasts are usually expressed in growth rates. They compound. Use a log measurement on the Y-axis of the growth graph. For example, news reports of the national debt of the United States emphasize how big it is and how fast the total amount is growing. Yet the compound growth rate has been fairly constant since 1980. The reason? Consistent behavior of both Democrats and Republicans – expand programs and cut tax rates.
(If you are an investor, be aware that the usual measures of risk or uncertainty underestimate them, even in the long run. You may sometimes read about “fat-tail” distributions. This means that extreme events happen more often than the distributions forecast.)
FORECASTS AND CORPORATE STRATEGIES
The present value of an investment or acquisition is usually sensitive to the long-term sales forecast. Even small changes in sales growth can make a big difference if sales growth is in percents and compound over time.
Why are long range forecasts important? Corporations make large investments in plant and equipment based on forecasts. Future cash flows, which includes profit, are discounted by the time value of money to determine whether or not the investment exceeds the company’s cost of capital. That, in turn, critically depends on the projected sales growth rate. My experience was that divisional managers who wanted the investment often made optimistic sales projections, far above the market’s expected long-term growth rate. They become angry when corporate staff killjoys like me redid their requests for capital at lower sales forecasts. Or, you can simulate the exercise at different sales growth rates to see how sensitive the discounted cash flow is to the sales forecast.
There is currently great uncertainty surrounding the future profitability of the massive investment in hyperscale data centers and AI platforms. Proponents point out the incredibly fast adoption and use of the AI tools. The danger is that the huge investment projected over the next four years may occur when sales growth starts to slow down. These are capital-intensive investments; their profitability depends on close to 100% capacity utilization. The danger is that the major players want to get their data centers up and running before the competition, without considering the future total industry capacity compared to sales.
Forecasting, econometric or otherwise, uses past data to predict the future. The base is to estimate the trend. Then other factors are added that are expected to change the trend and measure deviations from the trend. Decisions and actions are sometimes influenced by the forecasts.
One way to evaluate the sensitivity of a forecast is to assume a range of values for an important independent variable and run simulations. Change in demand for most products depends mostly on the forecast for the change in real income. The introduction of new products, or a whole new competing product category, makes forecasts less valuable.
If you sell globally, source globally, or have a global supply chain, this adds another dimension to forecasting and corporate strategy. Part of global planning is political risk of doing business in other countries. Foreign companies that invested in Russia after the fall of Communism got burned badly after Putin took over. Other considerations are exchange rates, tariffs and quotas, and cultural differences. Exchange rates have to be forecasted to determine hedging strategies. McDonald’s had a strict policy of no alcoholic beverages. But when they expanded to Europe, French franchisees insisted on serving wine and German franchisees insisted on selling beer. After some tense discussions, McDonald’s relented.
As a starting point, you could use an AI platform to search the internet for relevant forecasts of industry or national data. But again, approach with caution.
OUTSIDE INFLUENCES: THE EXAMPLE OF THE FED
The Fed says it bases its setting of the Fed funds rate, which influences many other short-term interest rates and indirectly long-term rates, partly on the Consumer Price Index (CPI) or some subset of it to measure inflation. Yet everyone, including Fed economists, know there are serious defects in the CPI. There is a minor industry of economists suggesting changes to the CPI. Fed policy based on the CPI may be ineffectual in reducing inflation or might even contribute to future inflation.
Why is this important? Demand for many products and services, such as housing and consumer durables like autos, partly depends on borrowing. Interest rates and size of monthly payments are often important variables affecting demand.
DEMOGRAPHIC EXAMPLES
For decades after World War II, when birth rates were high (the baby boom), no one expected birth rates to fall as fast or as far as they did. Then came the birth control pill, dramatic increases in the education of women, a rising percent of women in paid jobs, higher divorce rate, etc. Or life expectancies to increase as much as they have, thanks to innovation by drug and medical equipment companies. This complicated some short run and long run public decisions and plans.
Forecasts are inherently uncertain. Confidence intervals are important. Some long-term demographic forecasts contain confidence intervals. They can be very large. They are sensitive to small changes in the birth rate or life expectancies. For example, the Lancet population projections predict that the world’s population will peak at 9.7 billion in 2064. The confidence interval is 8.8 – 10.9 billion, about 1 billion each way. For 2100, the global population projection is 8.8 billion, with a confidence interval of 6.8 – 11.8 billion. The number for China’s huge population decline is from 1.4 billion now to about 730 million in 2100. The confidence interval is 460 million to 1.5 billion. This number is almost always quoted without the wide confidence interval.
For an individual country, changes in immigration policy can be important. In two years, President Trump has reduced immigration to about no net immigration. If this continues, the U.S., which was forecasted to have slowly-growing population until the 2060s, due to immigration and longer life expectancies, may start experiencing it as soon as the early 2030s. The problem for public policy, and the national debt, is that very slow or no growth population means growing numbers of senior citizens and a smaller labor force to support them.
Why study demographic changes? Because they influence macroeconomic growth rates, growth in income, composition of consumption, size of the labor force, and other economic variables that affect corporate strategy and performance. If you are selling to kids and not to senior citizens, your long-run sales could be in trouble.
THE PAST AND THE FUTURE
Much social, political, and economic commentary is a discussion of the conflict between the past and the future. Since the beginning of the Industrial Revolution, the past has represented continuity, the future has represented change.
The ability to adapt quickly to unexpected changes is a valuable management skill. A large difference between a forecasted variable and the actual number may indicate a reevaluation of assumptions, and maybe a change of strategies. Even better, start a company or work for a company driving the change.
The whole point of the Industrial Revolution is to disrupt trends and stability.
Even the largest, most profitable companies come under attack. A conservative portfolio of the dominant companies of the 1980s was a long-run investment disaster. Being in the S&P 500, a group of the largest public corporations, is now like being on a hit list because of rapid technological change. AI is accelerating that change.
One of the best management books is by Andy Grove, who made Intel a great company. The title of his book is Only the Paranoid Survive.
Long-range planning, including forecasts, may not always be valuable. But having managers taking part in the exercise is often valuable. It lengthens their time horizon with less concentration on quarterly results. It makes explicit how changes in industry or national data may affect corporate performance. And, hopefully, it makes management look over the horizon to see the coming tsunami.
Comments
Post a Comment