"They could be historic episodes [you factor in] or it could be things you just write down, and then you assign probabilities to those and you run them back," he says.
You also have to decide whether to show your clients a market that's serially correlated, uncorrelated or both, he says.
"Economists were comfortable for many years with the assumption that markets have no memory," he says. "In other words, that the probability of a higher or lower return next year is the same as the probability was last year-no matter what happened last year. Others, Jeremy Siegel [author of Stocks for the Long Run] would be sort of an extreme case, say, 'No, no no. If the market does really badly'-he won't say it this way, but I will-'If the market does badly it feels quite badly for those investors and it will work to do better next time.' If on the other hand, the market is giving you a lot of money it will say, 'Hey, they don't need it all the time. Maybe next time won't be as good.'"
While he says you can choose a positively correlated model like this, he argues that more economists would say it's better to assume that they are uncorrelated-that the probability of distribution of stock returns will be the same next year no matter what happened last year.
There's also a regime switching model you can embed in your assumptions, he mentioned, in which you can factor in a 95% chance the market will be normal this year, and a 5% chance it will come a cropper like it did in 2008.
Your job, says Sharpe, is to decide what models of the capital markets you're most comfortable with and which your clients should assume. That means choosing a set of assumptions, then running several different outcomes through a computer with your clients and seeing which of those disparate sets of outcomes they are most comfortable with.
"And believe me, the strategy you choose will depend radically on that particular set of assumptions," he says.
Sharpe suggests picking three different strategies-for example, a strict 4% distribution every year, or a managed payout fund from a company that resets the payout every year. To these strategies you can add capital market characteristics that you're most comfortable with, perhaps by using vendor software. You put the whole shebang through a Monte Carlo engine that will generate "zillions" of future scenarios with outcomes for the various salient factors. Then show the ranges that each of the three different strategies yields and find out which of those ranges a client is most comfortable with.
"In other words, focus the client on the outcomes-but on the range of outcomes, not a single outcome or a single outcome per year. Because as you know, that doesn't capture the reality."