I am a new member of my school's investment club. My sector is making a presentation in which we discuss what stocks we'd like to buy/sell. Part of this presentation is expected to use regressions. We have bloomberg data available for these regressions and I am no stranger to regressions, but I wanted to ask if anyone could provide me with information on the best regressions to run when deciding the value of a stock.

I'm assuming you mean creating an actual valuation model, DCF analysis and the works rather than just looking at market price?

Is it a dynamic model? As in, do you assign probabilities anywhere that are random, based on triangular distributions or cumulative distributions?

To make a regression I would suggest using VBA to run iterations, and create functions for t-stats, R2 etc. This would pretty much give you a sensitivity analysis.

Unfortunately I can only speak from experience and I have only made regressions for dynamic models.

Thanks for your reply. My finance ignorance prevents me from answering your questions but you have helped me get a better understanding of the situation.

Hopefully other members will be able to assist you better. Good luck!

You can use regression in one of two ways:

1) You can test the relationship between price and some multiple (P/E or P/B) to value the stock, or

2) You can use regression to estimate a key driver of the company's financials for an absolute valuation model such as DCF. For example, an industrials company is likely to see its sales tied to GDP. Likewise an Oil company to prices of oil. A regression of sales against GDP (possibly lagged by 1 to 2 quarters) might proivde you with a model to estimate sales which is the key driver in most valuation models.

Hope this helps.

This is not meant to be a hater post, but honestly I have never seen a sensible valuation based on a regression, ever. And I don't recall ever personally running a regression in six years on the Street. If you have to do it for an assignment, that's fine, but in the real world very few people do regressions for valuing stocks.

The problem with running a regression between price and multiples is that multiples change over time for any number of reasons as businesses evolve. Using historical multiples in valuation is both lazy and dangerous. Dressing them up with a regression is even worse.

The GDP example above does make sense, that would be one viable example, but don't even think about trying to estimate something down to the dollar -- it's going to be a messy, noisy estimate at best.

Try to avoid doing anything stupid like the person I interviewed last year -- I gave him a company and told him to come up with a buy / don't buy decision and the 2-3 most important questions he would ask management if given the chance. He told me that he had run a regression on the company's historical financials to estimate the fixed portion of the company's cost structure, and that his question would be why last quarter's operating margin was only 3.1% when his model said it should have been 3.5%. That was so unbelievably wrong that he got dinged on the spot (and then we laughed at him behind his back), and frankly it was pretty shocking coming from someone with 6-7 years of experience at an elite buy side firm. If you were to ask that question to management, any sensible, self respecting C-level executive would verbally punch you in the balls in the hope that you would no longer be able to reproduce, therefore saving the rest of the species from your stupidity.

Ravenous:

This is not meant to be a hater post, but honestly I have never seen a sensible valuation based on a regression, ever. And I don't recall ever personally running a regression in six years on the Street. If you have to do it for an assignment, that's fine, but in the real world very few people do regressions for valuing stocks.

Generally agreed, however, it is common on the street to use P/B vs. RoE regressions to value financial firms (back of the envelope). There are a whole host of theoretical statistical issues with this approach, but it is nonetheless a commonly used tool in many discussions.

Thank you for both of your replies.