Using Factor Models to Study Fundamental Drivers of Stock Prices

Summary

The document asks how to formalize a proposed relationship between an external time series, such as an industry or government statistic, and a stock's price. It also asks how to evaluate expected profit and risk if the external variable can be forecast. The question mentions Bayesian and causal graphical models but does not develop or compare them.

The accepted response points to factor models as a suitable framework for relating explanatory variables to asset returns, with prediction as a related application. It suggests that factor models address the general problem, while leaving the connection between factor models and graphical models unresolved. No model specification, estimation steps, validation design, trading rule, or performance evidence is provided here; the response directs readers to other discussions for those details. The practical lesson is to distinguish statistical relationships and forecasts from the separate task of assessing their trading profitability and risk.

Key ideas

• Factor models are proposed as a framework for linking explanatory variables to asset returns.
• Factor models can also be used as a basis for prediction.
• A forecast of an external variable does not by itself establish a profitable or suitably controlled trading strategy.
• The document does not explain how factor models relate to Bayesian or causal graphical models.

Tags

# How to formalize and validate models of fundamental factors involved price changes?


# How to formalize and validate models of fundamental factors involved price changes?












Suppose you have some stock X, and its price can be considered a time series. You believe that real-world number Y, like industry or government statistics, which is also time series, influences stock price.

How such kinds of influences be formalized and validated? Suppose you believe that you can somehow predict Y ahead of time, how would you evaluate potential profit and amount of risk involved?

In regular statistics, you may consider various graphic models, including Bayesian and causation networks. But that kind of analysis usually does not involve measures of profit and risk, as far as I know. You may say that estimator analysis should be related to risk measuring, but I am not aware of how exactly that could be used.

## Answer by uhbif19 (score 2, accepted)

https://quant.stackexchange.com/a/77265

After some searching around, I think that users of this site would probably say that factor models are the right formalism and these questions already cover mine:

- How to build a factor model?

- How to use factor models for prediction?

While I might specialize the question further on factor-model vs graphical model relation, overall I think it is probably not needed. So I would only close the question, not deleting it, to make it googlable if anyone would use words more closer to mine.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.