Rates, Agentic AI, Polymarket

Rates

Next week the Fed will almost certainly issue its first rate cut, likely 25bps, which will bring the rate between 5-5.25%. There are many takes on historical rate cut precedence, what has happened to markets in the past during rate cut cycles, but I am going to focus on just a few variables that I think are important or provide the most relevant context.

First, the 1 year treasury rate has dropped like a stone in water. The Fed sets the overnight policy rate, but the rest of the treasury rates trade on the open market. Yes, the Fed will buy or sell open market bonds with their balance sheet, or the Treasury will issue more or less of a certain duration, and those activities do influence the market rates. But, by and large, the Fed sets the overnight rate and the market sets the other rates. So what is the market saying? The 1 year treasury rate was 5.25% in April and it hit a low this week of 4%. The 2 year treasury fell from 5% to 3.6% over the same period. These are massive moves for a market like this. This tells us that we can expect short rates to fall pretty quickly over the coming eighteen months. Unless something changes dramatically, these market prices imply that the Fed rate will be near 4% this time next year.

The damage from higher rates has already been done and inflation has almost completely fallen back into what most would consider to be a “normal range.” The result of the rate hikes was the banks, commercial real estate, certain unprofitable long-duration equities and companies reliant on cheap debt got obliterated. Walgreens (WBA) was removed from the Dow 30 in 2020 and replaced with Salesforce. Part of this Dow swap was to change out legacy economy stocks with contemporary representation. Walgreens borrowed when rates were low and now they cannot afford to roll that debt over to higher rates. Their market cap, which was once over $100B is now closer to $7B. There are other issues with Walgreens, and this is not a comment on the stock, this is more a comment about where the rate hikes did their worst damage. The indexes are doing just fine, but there is considerable damage in many sectors and industries.

The best case for rate cuts from here on out is a series of predictable stair step 25 bps cuts at each meeting until the short end matches the rest of the curve. If by this time next year, the short rate is 4% and the 10 year rate is 4%, that would be a normal and reasonable outcome. It makes no sense for the short rate to be 5.25% and the 2-year rate to be under 4%. Cuts will also ease the interest rate burden on the US budget, which is painfully and absurdly high.

Agentic AI

AI agents are not a new concept. An AI agent is a multi-layer, multi-step AI automation tool. It’s essentially AI stacked on top of more AI. I think they first gained attention among the developer community about a year ago. Personally, I had not seen or used them until this past weekend when I saw that Replit was trending in the Apple app store. Replit is a coding platform popular with hobbyists.

I tried Replit’s AI Agent, which costs $26/mo. It works from your iPhone, but it makes more sense to use it on desktop. For example, it offered to build a stock analysis tool for me. First, it presented a roadmap of the features I might want and asked for my approval or modifications. It then built the entire application step by step, soup to nuts. The process was very cool to witness, but the results were mediocre. Based on its suggestions I asked it to build various websites with mapped data or products for sale. This prompted me to try other AI Agent platforms. They all produced essentially proof-of-concept work, not something fully useful or practical or ready for prime time.

AI continues to improve on a rapid iterative cycle. Just like the AI agents today, GPT-2 was mediocre, but ChatGPT was revolutionary. The trend suggests there will be a revolutionary AI Agent platform relatively soon, maybe next month or maybe next year. The problem with multi-layer agent AI today is that a single layer of AI often produces useless results, so stacking multiple layers of AI together only increases the probability of a completely useless output. Overall my experience with AI to date has been that it is beneficial in subtle ways. It’s hard to quantify the benefits, but they are real. I think the same is true for businesses. AI, in its many forms, helps improve products and reduce costs from inventory management to design work to communication. There is a rate of improvement baked in there somewhere that is real but difficult to calculate.

Polymarket

You may have looked at Polymarket during this election cycle. We’ve mentioned this in the past. It’s a betting market powered by crypto that continues to grow exponentially in popularity. Interestingly, you can see each user’s current positions and betting history.
Polymarket has quickly become an institutional-sized betting market. As of today, there are $850m in bets on the US election outcome. Some users hold over $3m on a Harris or Trump result. You can never trust the details behind any public information on any particular trade, but it’s still interesting.

Outside of politics, seeing other betting odds, especially those related to news or pop culture is interesting. This is $20m+ in total bets on the outcome of next week’s Fed rate hike. The 25 bps cut outcome is priced at 100/91, so risk 91 to earn 9 profit, while the “no change” outcome is 100/2, risk 2 to earn 98. The market expects a 25 bps cut, but the return on no change at 40:1 is pretty interesting. Typically, most people look at the futures market to evaluate Fed probability, so this is a useful second opinion.

I’d like to see the odds that Polymarket will soon face legal issues. The volume is getting far too large too fast to continue growing at this rate. Still interesting to watch this space develop.