The many faces of uncertainty

 

From the PowerPoint notes for the book Warren B. Powell, Reinforcement Learning and Stochastic Optimization: A unified framework for sequential decisions, John Wiley and Sons, Hoboken, 2022 (1100 pages).

This is a fascinating book that dives into the complex issues of stochastic optimization. At a high level, I found the description of many types of uncertainty beneficial. Only through dividing or classifying the many kinds of uncertainty can a researcher understand where the key risks are in a problem. 

Is there data uncertainty or model uncertainty? Is the inference uncertainty or transitional uncertainty? If researchers can explore these differences, they can better define the risks faced and the sensitivity to a given answer.

It would be interesting to take a given strategy and decompose all the uncertainties faced. 

The AI model used differs by task

 

I found this simple graph that matches how I use AI models. There is not one single provider that works best. The choice of AI model will differ by the task. Given these differences, I will sometimes use more than one model for he same task by asking the same questions or using the same prompts across several models and then comparing the results. Claude is easy to work with, but ChatGPT will sometimes provide more useful answers. Gemini and Co-Pilot are easy to use because of their integration with workflow, although their answers for more complex questions are not as clear.

While good at answering quick questions, their usage is harder to intergrate with normal quant work.

Bonds don't look cheap in current environment

 

Bonds are still in a bear market, but valuations still look cheap. The problem is determining the current fair value and whether there will be a move toward a new equilibrium. The simplest bond valuation will be real growth plus expected inflation plus any term premium. Expected inflation is above 2% and is unlikely to fall to the target rate. The real GDP is above 2% for the Blue Chip forecast and above 3% for the Atlanta Fed GDPnow forecast. Even if there is no term premium, we are looking at something close to the current 30-year Treasury yield and slightly higher than the 10-year yield.  

The only way to pick bonds is to assume a slowdown in GDP and a decline in inflation. This is possible for 2026, but this is not the current environment.

You are making a large macro bet if you think the bear market in bonds will reverse beyond current levels. 

 

Alpha and cost containment - The value of AI

 

We have written about how hedge funds are trying to contain costs by trading more efficiently. We are also seeing cost containment and efficiencies through the use of AI. Similar to consulting, AI can make analysts more efficient at some of their core tasks through summarizing and sifting through data in reports. The use of AI through EDGAR filing is not new, but has become a core part of the work by both discretionary and quantitative researchers. 

AI is being used as:

• information summary tool
• focused search tool 
• quick news analysis tool
• pre-screening tool along with quant analysis 
• simple idea generator
• proprietary prompt tool 

While not a research replacement, AI is not a research adjunct that allows hedge funds to run leaner shops with less costs on junior analyst development. The objective is to make senior analysts more efficient by reducing drudgery. Many firms have spent money on proprietary prompt libraries that can be applied to stock sets to serve as an alternative filtering mechanism. This can be especially powerful when linked with proprietary databases.