Hi guys,
I'm studying for CFA Level II at the moment and am covering the quants section of the course. Level II quants is mostly about using regression analysis. I find the whole process interesting and it is something I would like to learn, however I have yet to see anyone actually use any of this stuff on the job.
I am a trainee corporate credit analyst and most of what I do revolves around forecasting financials and computing ratios. There are a few examples where I think a regression model could be used to forecast sales, such as for oil & gas companies with regards to oil prices, however beyond that I think the scope is limited.
Is there anyone in the industry who could comment on this or shed some light on their own experiences? Is it worth my while devoting time to learning this stuff or is it a load of fluff?
Thanks!
time series analysis and regression are complex and not reliable.
don't use it or have an expert to do it
We use regression analysis all the time.
Simple OLS it's ok for most simple stuff (assuming you know how to read the outputs and understand if the model makes sense).
For pair trading the best is to learn co-integration.
For advance electricity pricing I would recommend learning about regime-switching (this one is a pretty advance topic)
Happy to help with any question you may have regarding this stuff.
We use it all the time too.
As an example, the primary driver between one of our covered companies is the mix between product segment X and Y. So I ran a regression on change in mix to consolidates GP (R^2 = 95%+). Now, I can model the growth of each product line separately and have gross margin as an output. This effectively reduces my margin of error.
lol
What about the Durbin-Watson? People tend to look at the p-values (for variables and for the overall model), Adj. R-Squared but tend to forget the DW.
If I'm doing something wrong, just let me know.
Granted that the sample size is small, but I calculate a correlation between residuals of 0.22. So 2*(1-0.22) = 1.56. Dl and Du are about ~1 and ~1.3, respectively.
But I've used this method in combination with qualitative factors to forecast future GMs and I've been accurate within 10bps.
http://i39.tinypic.com/qrxamv.jpg
wow... at a lackj of words
Level 2 candidate here as well...
I do equity research for a hedge fund.
I've used regression analysis a few times, particularly when a company (let's say retailer) pre-announces sales figures. You want to get EPS and can make educated assumptions about margins, whether OpEx goes up or down, etc.
I've found regression useful in supporting my assumptions - Marketing Expense growth is 85% correlated with sales growth, for example.
I'd never use it for forecasting though. Maybe in a world without management commentary or where all you are given is an income statement data set.
I used it in IB all the time. Specifically to estimate what a client company's newly issued debt would be rated in comparison to it's peers based on a number of multiples and metrics.
I've used regression analysis to value a company. You take a comp set's ROATCE and their P/TBV come up with a regression line. Plug in your targets ROATCE and you come out with a price.
ARCH and ARIMA for forecasting
Plenty of basic OLS is used in sales forecasting
Personally, if a simple regression works well (with good p-values, Adj. R-squared and DW) I prefer to use it. I tend to see GARCH more as a way of getting rid of volatility clustering.
Sure. I always try to start with as simple of a model as possible. Problem is almost all of what I work with is going to have autocorrelation or volatility clustering as you mentioned out the ass.
Tough luck man :)
I was just going over two stock series and although the simple regression seems to work well there is volatility clustering, so GARCH it was.
Thanks for the affirmation. I didn't think I was doing anything blatantly wrong either - I guess others are quick to assume/judge.
I'm a numbers guys but I don't use math as my safety blanket! because I don't believe one can simple model the future based on historic trends. That is why I added the phase "combined with qualitative factors."
The model atop that I posted is not a simple "change in mix model." I did a little more work with the data - but that's not for public consumption.
Non cumque corrupti illum amet. Incidunt consequatur accusantium optio minus. Et laboriosam temporibus reiciendis. Sed facilis quam magni. Et sint corrupti enim et sed corporis et consequatur. Necessitatibus voluptatem qui natus ducimus est in qui. Et ea eos harum necessitatibus quia sunt consequatur.
Tempore vitae qui soluta corporis nobis soluta odit cupiditate. Quam esse ut excepturi exercitationem accusantium rerum quos. Dolorem quas ut autem nulla sed. Itaque inventore repellendus mollitia facilis.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...