How to get better at forecasting.
I have found neither of these methods to be all that helpful. The informal process lacks a rationale that can be used to guide decision making. And with little documentation, an informal process makes adjusting the current forecast and improving the accuracy of future forecasts difficult.
On the other hand, the formal process gets too detailed, losing the forest for the trees. With so many inputs, the current forecast is also difficult to adjust. And although the accuracy can be precisely tracked and variables adjusted, the formal process often strives for an unnecessary level of accuracy.
MAKING BETTER DECISIONS
In order to design a valuable forecasting process, the objective needs to be clearly stated. My objective for forecasting is to understand the economic cycle and its impact on my business so that I can make decisions that will maximize profits. When I create a specific forecast, it is just as important to understand the direction the business is likely to go and what is the stage of the economic cycle. It is very difficult for any individual business to move in a different direction than the industry it is in. But understanding the cycle can help a business make better decisions regarding when to invest in capital equipment, hire new employees, develop a new product, and more. A business with economic understanding will reach higher highs during peak business times and higher lows during slow business times.
Figure 1 (from Ahead of the Curve by Joseph Ellis) shows the most important data points in order to follow the economic cycle. Each box represents a different data point tracked by an industry association or government office. (You can find most important economic data at http://research.st1ouisfed.org/fred21.) The solid black lines indicate that the data point in the preceding box is a primary leading indicator for the data point in the following box.
For example, interest rates are a primary leading indicator for consumer borrowing. The dotted black lines indicate that the data point in the preceding box is a secondary leading indicator for the data point in the following. For example, jobs are a secondary leading indicator for consumer borrowing. So, while both interest rates and jobs can give us an idea of what will happen in the future regarding consumer borrowing, interest rates are more important than jobs in forecasting the future of consumer borrowing. The flow of the economic cycle moves from left to right on the chart. So, inflation and income are some of the earliest leading indicators in the economy, while capital spending and jobs tend to be lagging indicators.
In addition to these data points, I use my own company data, private research, and surveys to create my forecast. You can also include information gleaned from your networks, although I try to rely mostly on the various data points in the economic cycle, so that I'm not influenced by the opinion of others when creating my forecast.
CHOOSING DATA POINTS
When creating my forecast, I don't use all of the data points in Fig. 1 in addition to my own company data. Instead, I try to pick just three to five data points that move left to right on the chart so that I can create a series of leading indicators for my business. I would rather have too few leading indicators than too many. It's easy to become confused when you have too many leading indicators to follow.
Once you have chosen the data points that correlate best to your business, you need to put them in the proper context. There are several important aspects to creating the proper context for the data in your forecast.
First, you should present the data in a graphical display. Most of us are not very good at picking up trends by looking at a table of numbers, but all of us can tell at a glance if a line is moving up or down.
Second, your charts should include two correlated data series so that you can see how the two data series move in relation to one another.
Third, your graph should cover a significant period of time. All of my forecasting charts span a 40-year period, a length of time that ensures I get at least three or four complete economic cycles. Typically, you will have 10 or 12 cycles for the two data series you're correlating.
Fourth, you need to use real data, not nominal data. So, if you are using data that is captured in dollars, you need to adjust for inflation. When possible, I try to find data that is counted in units other than dollars.
Filth, use rate-of-change curves to more easily see the cyclical fluctuations and correlations between the two data series.
CALCULATING RATE OF CHANGE
Although they may take some getting used to, there are a couple of reasons for using rate-of-change curves. First, it is easy to see the momentum in a particular data point. A rate-of-change curve tells us whether or not the data point is growing or contracting and how fast it is growing or contracting. Second, it is easier to see the correlation between the two data points. Third, it is easier to see the time period between changes in the leading indicator and the data point you want to forecast. When the data is presented in rate-of-change format, you can easily see how the peaks or troughs in one data series lead to the peaks or troughs in the second data series. Then you can estimate the time lag between the peaks and troughs between the data series. This time lag is important because it gives you a reasonable approximation of how many months out you can reliably forecast when using rate-of-change curves.
Such curves are typically calculated for one-, three-, and 12-month periods. One-month rate-of-change curves are very noisy. That is, the curve moves up and down a lot, and the overall trend is harder to see. But it is a more sensitive indicator. Twelve-month rate-of-change curves are much smoother. Therefore, it is easier to correlate two data series using 12-month rate-of-change curves. The downside is that they are not as sensitive. I calculate the rate of change for all three time periods for all of the data I track, but I only use the 12-month rate of change on my forecasting charts.
Calculating rate of change is pretty easy. To calculate the one-month rate of change, divide the current month's data by the data from the same month one year ago. For example, to calculate the one-month rate of change for July 2013, divide the data point from July 2013 by the data point from July 2012. Then, to put that number in percent terms, simply multiply it by 100, then subtract 100. Three- and 12-month rate-of-change curves work exactly the same way, except that instead of using the data from just one month, you are using either quarterly or annual data. So, to calculate the 12-month rate of change for July 2013, you would first need to sum the data points from August 2012 to July 2013 and August 20:11 to July 2012. Then divide the first sum by the second sum. Then convert it into percent terms just like for the one-month rate of change. Really, it is very simple math.
UNDERSTANDING RATE OF CHANGE
Also, it is important to know how to read a rate-of-change curve. Figure 2 shows a sample 12-month rate-of-change curve. It is always important to identify the zero line on a rate-of-change curve. This is because anything above zero indicates growth, regardless of whether the line is moving up or down. And, anything below zero indicates contraction, regardless of whether the line is moving up or down. The further the line moves away from zero, the faster that data series is growing or contracting.
The four arrows on the chart indicate the four parts of the economic cycle. When the line is above zero and moving up, that is accelerating growth, which means that data series is growing faster and faster each month. When the line is above zero and moving down, that is decelerating growth, which means that the data series is still growing but it is growing more slowly each month. When the line is below zero and moving down, that is accelerating contraction--the data series is contracting faster and faster each month. Finally, when the line is below zero and moving up, that is decelerating contraction. The data series is still contracting, but it is doing so at a slower rate each month.
FORECASTING THE PLASTICS INDUSTRY
To forecast the plastics industry, I start with interest rates and housing permits (see Fig. 3). Remember from our chart of the economic cycle that interest rates lead housing permits. For the interest rate. I use the real federal funds rate, but you can also use the 10-year Treasury bill rate. For interest rates, don't calculate rate of change. Rather, simply determine the basis-point change in the rate from the current month to the same month one year ago. Because falling interest rates boost economic activity, I have put the negative change in interest rates at the top of the chart. Based on this chart, by looking at the time lag between the peaks or troughs in both data series, it appears that changes in the real federal funds rate lead changes in housing permits by one to two years.
My next chart correlates housing permits with plastic products industrial production (Fig. 4). This chart shows that housing permits tend to lead plastic products production by six to 18 months, with an average of about one year. This mean., that based on the trend in housing permits, you could forecast the next 12 months of plastic products industrial production.
However, instead of plastic products industrial production, you could use the number of parts produced by your own business or the dollar volume of your orders (remember to adjust for inflation). My guess is that you would see a similar correlation--about one year--between housing permits and your production or orders.
FORECASTING YOUR BUSINESS
You might be producing parts for a specific industry. In that case, instead of using housing permits, you might want to use data specific to that industry as a leading indicator for your production. or orders. For example, if you make parts specific to the automotive industry, then you could use motor-vehicle and parts spending as a leading indicator for your production. Good leading indicators for motor-vehicle and parts spending are interest rates, consumer borrowing, and personal income.
I've been using this forecasting method in my business for more than five years. It has been a pretty easy way to forecast 12 months out with [+ or -] 10% accuracy. In fact, Lam able to produce more accurate forecasts than our brand managers, who have industry knowledge and understand the relationships with our customers. As you apply this method to the trends in your business (e.g., is October usually a better month than March?) you should be able to generate relatively accurate forecasts reasonably far in advance.
One final point: I keep track of every forecast I create, along with any adjustments 1 make. This allows me to see whether my tendency is to be too pessimistic or optimistic heading into or out of recessions and other changes in the economic cycle. This should make my forecasts more accurate as time goes on.
ABOUT THE AUTHOR
Steven Kline Jr. is part of the fourth-generation ownership team of Cincinnati-based Gardner Business Media, which is the publisher of Plastics Technology. He is currently the company's director of market intelligence. Contact: (513) 527-8800: email: email@example.com; blog: gardnerweb.com/economics/blog.
By Steve Kline, Gardner Business Media