Hi-Def Projection
Hidden in the accounting, a more accurate way to forecast sales
The conventional way to forecast retail sales is more scientific than a tea leaf reading, but not by much.
Even as data abounds, the abiding method boils down to a surprisingly simple extrapolation: take last year’s growth rate and project it forward.
Now, researchers at the Universities of Washington and British Columbia have developed a significantly better forecasting tool. Their model incorporates previously overlooked public accounting disclosures to produce a forecast of sales—a strong indicator of future firm value—that is twice as accurate as conventional models.
The new technique factors in a firm’s projected store openings. More importantly, it computes expected sales for each based on the firm’s historical new store sales figures.
“Our model is a simple and yet powerful way to forecast a firm’s future sales based on publicly available data,” says Sarah McVay, an associate professor of accounting at the UW Foster School of Business. “Even a small improvement in forecast accuracy can consequently have large implications in the accuracy of value estimates.”
Variable sales
McVay and co-authors Asher Curtis and Russell Lundholm developed and tested their forecasting model on 87 retail firms, using sales data from the years 1995 to 2010.
Their key innovation was including a firm’s projected new store openings in the coming year, a nearly universal disclosure in retail financial statements. Any new store automatically adds sales. But how many? New stores perform differently for different retail companies. Differently, but predictably.
To define those predictably variable sales—and refine the forecast—the authors needed to look backward. Analyzing hand-collected sales data, they were able to calculate each firm’s average annual sales per new store, as compared to established stores.
A new Costco, for example, generates enormous excitement and nearly two times the sales, in its first year, as existing Costco stores. A new Starbucks café, on the other hand, takes time to ramp up to the level of established stores, and generates only half the sales in its first year.
Knowing the relationship between new and established stores allows a forecaster to generate a superior forecast. Tested against traditional methods of forecasting sales, the new model cuts the rate of error in half, from four to two percent.
“What distinguishes this model and makes it useful is that we can break apart the old from the new and recognize that they’re going to generate different sales rates, but in a predictable way,” says Curtis, an assistant professor of accounting at Foster.
A working model
For any model developed in academia, the proof is in the practice. Unlike most models rolled out in scholarly journals, this one is already being used by retail industry analysts—whose livelihood depends on the accuracy of their forecasts. After all, sales is a leading indicator of a firm’s market value: prized information for investors the world over.
After Lundholm, a professor of accounting at UBC, presented the paper at several investment banks, their analysts immediately adopted the new model into their calculations of future retail sales.
“We find systematic evidence that these sophisticated analysts are increasingly incorporating our information in their projections,” McVay says. “It’s enormously satisfying to see our research directly affecting practice.”
Extendable logic
McVay and Curtis say they built their model around the retail industry because so much is known about new store plans and per-store sales. But they say that their model could be adapted to forecast sales in any industry. Past data could be used to calculate the impact of adding a new ship to a cruise company’s fleet, for instance.
“Our model illustrates a structured way to handle information about future sales-generating units and the future sales rates per unit and thus is applicable to virtually all industries,” says Curtis.
Moreover, the logic of their model could apply to forecasting sales of new products—and their impact on a firm’s balance sheet. So knowledge of plans to launch a new mobile phone or board game or vacuum cleaner, coupled with analysis of the manufacturer’s past launches of similar products, could produce a more accurate forecast of overall sales.
“You could certainly think about how new products perform versus old products,” McVay says. “You couldn’t use the exact model. But the logic should hold up.”
“Forecasting Sales: A Model and Some Evidence from the Retail Industry” is published in the summer 2014 issue of Contemporary Accounting Research.
