What a financial product promises determines how it will be judged, and also how it will be designed. A good promise meets three conditions: the market can provide it, the product itself can deliver it, and the user can verify it. Excess returns and the direction of prices fail the first condition; short-term returns fail the third, because they are mainly a matter of luck. What can be promised is cost, execution quality, adherence to the agreement, coverage of ongoing tasks, and results and processes that can be checked. A promise should also be set in terms of net value, counting the user's effort and the risk borne, and written out separately for each task the user wants done. What most easily damages trust is using selected backtests and rankings to create expectations that cannot be met.
A promise must pass three tests. First, can the market provide it? Time compensation and risk compensation are available to everyone; excess returns sum to zero overall, and the direction of prices is hard to predict. To promise either of the latter two is to promise something the market cannot give. Second, can the product deliver it? Fees, execution quality, rebalancing discipline and adherence to constraints are all within the product's own control; returns are not. Third, can the user verify it? A promise that neither the user nor a third party can check is only advertising. Few promises meet all three conditions, but it is exactly these promises that make up the value a product can sustain over the long run.
Even if one product really is better, its short-term returns may lag. Holmström (1979) showed that an agreement should rest on the signals that best reflect the agent's effort; returns contain too much luck and are a poor signal. Under the illustrative assumptions in the figure, the better product earns 1 percentage point more each year and the annualized volatility of the difference between the two is 10%; judged on one year of returns, the better product wins with a probability of only about 54%. Process measures can be checked every month and carry little noise: whether fees stayed within the promised range, whether the portfolio drifted from the agreed allocation, whether every cycle of an ongoing task was accounted for. With them, good and bad can be told apart almost every time. Promising the process lets users verify it and also aligns the direction of the product's effort with the user's interest.
Judged by returns, the better product often looks worse
Product A really is better than B: 1 percentage point more each year (illustrative). Drag the evaluation period and compare, under two ways of judging, the probability that A is correctly judged better.
Consider two products. One advertises higher returns but requires the user to watch the market every day, make frequent decisions and bear risks that are not made clear. The other has ordinary returns but takes almost none of the user's time, and its risk stays within a range stated in advance. Comparing returns alone, the first is better; once the user's time and the risk borne are counted, the conclusion may be reversed. French (2008) estimated that from 1980 to 2006, the management fees, expenses and trading costs that US investors paid in pursuit of excess returns averaged about 0.67% of the market value of stocks each year, and bought no excess return in aggregate. A product's promise should be stated in terms of net value: what the user gets, what the user pays and what the user bears.
Different tasks need different promises. For looking up a figure, the promise is accuracy and timeliness, with the source and the point in time noted. For researching a company, it is that the evidence can be traced and that the conclusion states the conditions under which it would be overturned. For ongoing monitoring, it is that every cycle completes the checks within the agreed scope and distinguishes "no change" from "check blocked". For managing a portfolio to a goal, it is that the agreed goals and constraints are always observed, that departures are referred to the user, and that execution is at reasonable cost. Each promise corresponds to a measure that can be checked, and to a way of checking it.
Writing promises task by task
Choose a kind of task to see what the product should promise, which measure it is judged by, and how it is checked.
The most common form of overpromising is to show an attractive historical backtest or a short-term ranking. Bailey, Borwein, López de Prado and Zhu (2014) showed that if enough strategies are tried on the same stretch of historical data, the best performer will look excellent even when none of them has a real edge; the more strategies tried, the better the best backtest looks, and it has almost no power to predict the future. The marketing rule for investment advisers that the U.S. Securities and Exchange Commission revised in 2020 also sets strict conditions on presenting hypothetical performance. A product may show history, but it should state how many variants were tried, on what sample they were tested, and how they performed out of sample.
Promise only value that the market can provide, the product can deliver and the user can verify, and state it in terms of net value and by task.
Fees, execution, discipline, constraints and coverage can be promised; returns and direction cannot.
Write measures such as fees, allocation drift and accounting for each cycle into the promise, and publish them monthly.
Say what the user gets, pays and bears, not just the returns.
Do not use selected backtests and short-term rankings to create expectations; write out, task by task, what is and is not promised.