Trading with signal and price impact uncertainty

 

When you want to make your model practical, you have to look at signal uncertainty and price impact. If you use any model, your one-step-ahead forecast may have some uncertainty because the average coefficient may not accurately represent the true coefficient sensitivity at a given time. Similarly, for any model there is going to be a prcie impact from trading. If the signla is weak then iapct uncertainy is not that important but as the siganl strengthens, the effect price imapct will become more important. See “Trading with uncertainty about signals and price impact.” These real-world effects are important if you want to trade a model.  

Expectation Bias and short-term Momentum

 

Machine learning can be used to predict analyst forecast errors. These forecast errors can predict cross-sectional returns and abnormal returns. There is an underreaction to fundamental news, which is amplified by overconfidence, sticky beliefs, and information uncertainty. This expectation bias can explain short-term price momentum in high volatility stocks. From expectation bias comes a rationale for trading patterns in price. See "Expectation Bias and Short-term Momentum". Past forecast errors have a critical impact relative to other features.

Attention versus earnings and momentum

 

One area of increasing research is the attention that is given to a specific market. Investors cannot follow everything, so there are different levels of attention. Given changing attention, there should be different levels of efficiency. What is found in the paper, "A Tale of Two Anomolies: The implications of investor attention for price and earnings momentum," is very interesting. Stocks that show more attention show greater price momentum and weaker earnings momentum.  The authors make the following claim: investors pay less attention to earnings news. stock price underreacts, leading to stronger earnings momentum, but when attention is high, behavioral biases intensify, which fuels overreaction and price momentum. Depending on the type of attention, the effects will vary. 

Multi-agent LLM systems for profit

 

LLM can be used to develop autonomous trading systems. The key is to have agents mimic the workflow within a multi-agent trading framework that divides a trading system’s tasks. When you look at a deep decomposition of the tasks associated with a trading system, you will get better risk-adjusted returns. In the paper, “Toward Expert Investment Teams: A Multi-agent LLM System with Fine-Grained Tradign Tasks”, the researchers develop what they call a fine-grained system that looks at detailed tasks and instructions and compares it with a coarse-grained LLM system that does not provide detailed instructions to the agents. 

Across different stock portfolios, the fine-grained approach performs much better. It is notable that we can examine the effects of dropping specific agent tasks and assess their impact on the Sharpe ratio.

The results are clear - be detailed with the LLM tasks given to agents. The more specific you get, the better the results.