Amtrak's models use a very large number of independent variables to refine a basic forecast derived from history, Stein explains. "We have substantially increased the sophistication of those models in recent years. We use core patterns that we adjust for the multitude of special items and events that we know about."

Amtrak's cash forecasting and hedging operations are not joined at the hip. But shared data means risk management and cash management are definitely coordinated.

Amtrak uses three models, Stein says, one for revenue, one for expenses, and one managed directly by treasury that that bridges the gap from GAAP-based financial statements to cash.

"We make sure that all the non-cash items we know about are reflected in the forecast directly or converted to a cash basis," he explains. "And we forecast all the changes to the balance sheet that are occurring, whether or not they show up in the income statement."

The revenue model incorporates such things as the health of the economy and changes in employment and disposable income by region, as well as fuel prices. When there is a change in service–like an increase in the number of Acela train runs between Washington and Boston each day–Amtrak has a good idea how that will affect revenue and forecasts accordingly. Any significant reduction in trip time is likely to mean more passengers and revenue, Stein adds. If it's leap year, that makes a difference. "All this is quantified and fed into the forecast," Stein says. The expense forecast is simpler because Amtrak operates with a lot of fixed costs.

Then treasury takes the revenue and expense forecasts and factors in balance-sheet items that aren't part of the P&L, like capital expenditures, Stein explains. "We ensure that the data are consistent, both mathematically and conceptually. And we calendarize what we know into monthly forecasts for each major account on the balance sheet." Each sub-forecast is refined by repeated regression analysis. "That's the key to converting historical results into a good forecast," Stein concludes

Amtrak's forecasts depend on hard work more than slick technology. "I get a 10-page report daily that breaks down all the cash flows from the prior day," Stein says. "It shows me deviations from the forecast, and I can immediately drill down in the prior day report of cash movements to see where the deviations came from. We always know how we're doing. And we always know why."

And the railroad's treasury works hard at continuous improvement. "We go through the monthly forecasts every month, making line-by-line comparisons between what actually happened and what we forecast," Stein says. "We look into what caused each variance. If a contract was signed in February when we had expected March, there's not much we can do about that. But we look for flaws in the model and ways to improve it–like if we'd assumed a three-month lag between recognizing revenue and actual disbursements and it was happening in two months. The models are changing all the time as business goes on. They are very dynamic."

Such elaborate forecasts do not mean that Amtrak has acquired prophetic powers. "In a steady state, the models work pretty well and the variances are immaterial, but we don't get very many steady-state days any more," Stein says. The payoff isn't being right but being prepared. "Good forecasting allows us to anticipate the fallout of unexpected events and take effective action quickly," he says.

Forecasts are also improving at $1.3 billion AARP, also based in Washington. "Since we built our forecasting system five years ago, our forecasts have become more sophisticated," says Bret Hom, director of treasury operations. "We keep reaching out to more departments for input, and we keep enlarging our list of top vendors and revenue sources that we track on an individual basis. Using actual-vs.-forecast scorecards is helping us find where we are off and improve. Our forecasts have definitely become more accurate."

As a not-for-profit service organization, AARP has no cash conversion cycle and depends more on forecasts to anticipate miscellaneous revenues and expenses. "Because of our unique disbursement stream, we can't find a commercial product that will work for us, so we build our own," Hom says.

That system zeroes in on AARP's largest vendors and revenue streams. For example, in May it was experiencing a seasonal spike in reimbursement expenses for volunteers who help elderly people with their tax returns, he notes.

Once companies gather the relevant data, how it's organized is critical. For longer-term forecasts, treasuries often build forecasts around buckets–one for booked transactions, one for contractual or firmly committed transactions, one for likely but not yet committed transactions and one for contingent transactions, explains Ron Chakravarti, managing director and head of liquidity and investments solutions for Citi Global Transaction Services. "They may have different policies and procedures for each bucket."