Source: Perplexity.AI

Here’s a bright—but bad—idea for how to use artificial intelligence to boost your investment results. It illustrates the potential and the peril that we face as AI infuses our lives.

Let’s say you are the treasurer of a city in Texas that has eight competing LGIPs. Or California where there are five. It would be straight forward to commission an AI agent to keep track of your pool balance, check the yield on each fund every morning and move all funds to the one that posted the highest yield. Like magic, no human intervention and you could expect to  pick up a few basis points of yield.

If the pool portals you use would not permit an AI agent to move money, you could assign an entry level human employee to implement the AI agent instructions. It might take 10 minutes a day, almost as efficient as having the AI agent do it all.

Pretty simple, right?

Well, this is both a dumb idea and a bad one.

It’s dumb because for the vast majority of stable value LGIP portfolios “yield” is a backward-looking number representing the income an investor would have received over the prior seven days. No doubt there are LGIP investors who spend time yield-shopping, but moving money for a few basis points of “yield”  is dumb.

That’s not to say yield should be ignored. If you compare fund yields or compare them to a benchmark like a PFII LGIP Index, an S&P LGIP index of rated LGIPs or a Cranedata index of money market funds over a period of time—say six months—you may identify consistent differences and it’s important to understand why. It’s the “why” that should go into your investment decision process.

The first place to look at why  is fund expenses because they are deducted from income to determine yield. LGIP portfolios have widely different expense ratios, varying from a few basis points to 30 or more. They also have different services and different levels of risk.  Differences in expense ratios may explain most or all of observed differences between one fund’s yield and that of another.

If a fund has been tracking consistently to a benchmark index and the tracking suddenly changes that should trigger a question and review. And it’s not dumb to have your AI agent keep track of this for you. If the benchmark is an index of LGIPs or money market funds, any change in the variance of more than a couple of basis points should raise a flag. Maybe the fund manager has temporarily changed the expense ratio. Maybe the portfolio manager has made a particularly good bet, or a particularly bad one. Don’t hesitate to ask and consider the answer before you act.

It’s not dumb to track yield closely, but it is not smart to have yields drive your investment decisions.

Now to the bad idea part.

Much of the concern about AI focuses on whether it will displace human work or amplify the efforts of hackers, but there is another relevant aspect:  its potential to disrupt and destabilize financial markets in ways that we are only beginning to comprehend.

LGIPs are in the business of intermediating funds, just as money market funds and banks are. Stable value LGIPs convert cash which is immediately available for use to short-term securities that mature in 30 days 90 days or even one year. The LGIP then collects the difference in the time value of money that it pays to shareholders. This intermediation creates a mismatch of liquidity and maturity, and it works as long as the LGIP investors don’t all rush for the exists or flood a fund with money at the same time. The problem is that an AI agent can accelerate the movement of money to the point where it is destabilizing.

The prudent manager of an LGIP will evaluate past investor behavior, study cash flows, and keep enough funds liquid so that even an unlikely event—say a big move in rates—will not exhaust its liquidity. The prudent manager may discourage hot money investors to dump large amounts into the fund to take unfair advantage of rates. But investor inertia is also a factor that affects fund flows. Some investors are quick to respond to market signals while others are not. And investors react to the signals differently, so a change in market rates might motivate one to move immediately while another decides to wait for further information or even to prioritize something other than moving funds now.

But what happens if investor decisions are performed by an AI agent that is prompted to move all the investor’s balances into or out of a fund immediately once an event (say two basis points of yield differential) is observed?  Cash flows into and out of the LGIP would be compressed in time; the fund’s ability to intermediate between investor/shareholders and investments would be severely compromised.

I’s possible to imagine that AI-driven investing designed to chase a basis point or two could result in swings in LGIP balances of 20% or 30% or more in a day. That would amplify fund asset volatility, force a fund’s manager to adopt a radically different investment strategy, and reduce the benefit that funds are able to provide investors.

From this perspective AI-driven investment is a bad idea. That’s not the only reason major banks have blocked AI agents from interacting directly with them. They worry about AI agents that successfully hack an account and impersonate the real owner, AI agents that hallucinate, and AI agents that simply do the wrong thing.

Bottom line:  Maybe a frontal assault by AI agents will not happen, but for all the potential of AI there are risks, some known and some that we will understand only after they’ve inflicted something bad. The challenge for public funds investors—indeed for all of us no matter what our roles—is to consider the risks of this emerging technology before we apply it.