A Bot on Wall Street

I · The Screen With Two Rates

A New York investor opens two browser tabs. In one is a bond fund’s yield, printed to two decimal places. In the other is the rate on a margin loan secured by a portfolio. The investor imagines borrowing fifty thousand dollars, buying the fund, and letting a script watch the difference. It is the kind of idea that becomes more persuasive when put in a spreadsheet: one column for income, one for interest, and a green cell at the end. The idea is hypothetical; the question it raises is real. What would the green cell have left out?

The first omission is that the two rates are different kinds of numbers. A fund’s thirty-day SEC yield approximates income earned by its holdings over a past thirty-day period, after expenses, annualized against its price. It is a standardized disclosure, not the rate promised to the next buyer. The broker’s margin rate is the price charged on a loan and can change under the account agreement. Neither rate freezes the market value of the fund. A script can read each number faithfully while mistaking a snapshot for a contract.

For a clean illustration, suppose the fund shows a yield of 4.75 per cent and the dollar loan costs six per cent. On fifty thousand dollars, the first number suggests $2,375 of annualized income; the second implies $3,000 of annual interest. The difference is minus $625 before trading costs, taxes, or any change in the fund’s price. These are assumed rates, not a quotation from a broker or a forecast for a fund. The calculation is useful precisely because it needs no model to expose a losing starting position.

II · The Other Number

There is a New York rate that will be tempting to substitute for the investor’s loan rate. The Federal Reserve Bank of New York publishes the Secured Overnight Financing Rate, or SOFR, a broad measure of overnight cash borrowing backed by Treasury securities. It is a benchmark drawn from wholesale transactions. It is not the price a retail brokerage customer is offered to borrow against a mixed portfolio. The difference between those two numbers is part of the business, not an inefficiency that a home-built program can collect.

One can compare a fund yield with SOFR to ask a market question: how much extra yield appears to be paid for duration, credit exposure, and other risks? To ask whether this investor has positive carry, one must compare the fund with the actual all-in loan charge in the actual account. If the spread is negative on that basis, faster polling only reports the loss more promptly. If it is positive, it is still a spread between a historical income measure and a floating financing cost, with the price of the bond portfolio left outside the arithmetic.

III · Whose Clock Runs First

The second omission is control. The script may be told to sell if the loan-to-value ratio approaches a limit. The lender has its own rules for what the collateral is worth and how much equity the account must maintain. Those rules are not subordinate to the customer’s code. The Securities and Exchange Commission warns that a broker may sell securities in a margin account without first consulting the customer when the account falls short. Interactive Brokers’ risk disclosure likewise says it generally will not issue margin calls and may liquidate positions without prior notice.

A daily check is therefore not a daily guarantee. Bond prices can move before the script runs; the broker can change a house requirement or act at a time the script did not choose. Even a flawlessly programmed sale may come after the broker’s sale. The program can record the event, but it cannot reverse the fact that borrowed money magnifies the consequences of a decline in the asset bought with it.

The third omission is the difference between income and return. The yield number says something about income in a recent window. A total return also includes changes in the fund’s price. A fund can distribute cash while its shares fall by more than the distribution. The loan remains a dollar obligation in either case. It does not shrink because the asset pledged against it had a bad month.

IV · A Useful Machine

There is still a job for the program. It can collect the published yield, the actual loan charge, the value of the collateral, and the broker’s current margin requirement. It can log every change and warn the investor when an assumption has stopped holding. It can insist that a forecast distinguish a distribution from a gain and that every claimed spread be calculated after financing. These are valuable forms of memory.

But an alert is not an execution right, and a backtest is not a lending agreement. The investor on the other side of the screen must still decide whether the prospective return pays for the risks the loan creates, including the possibility of an involuntary sale. In New York, where a published benchmark can feel like the going price of money, the most useful line of code may be the one that refuses to substitute the benchmark for the bill.

Revised for New York with AI on October 12, 2026; the sources above are primary disclosures. ◆