Hello, Throughout the session, I have seen that series are changed into log ones. Why is this better? If your data cannot become negative, why should it be transformed? Thanks for the help!

They’re creating linear models. If you have data that grow exponentially, a linear model on the original data is inappropriate, but a linear model on the logarithm of the original data is appropriate.

In general, there are a bazillion types of transformations you can make to data to make the resulting relationship (approximately) linear. Because some financial data tend to grow exponentially, the CFA curriculum focuses on logarithmic transformations.

CFA stock returns concepts are based on reversion to mean. That is a security cannot go to high or too low as over time it will revert to mean. if you want to project returns of a particular stock, it’s not possible to project exponential return neither substantially negative returns. In real life, though some stocks do give exponential returns but that is not based on fundamentals more on market speculation. Think of bitcoin as an example.

This may be true, but it’s not related to the original question.

Returns aren’t exponential.

If returns are constant, values grow (shrink) exponentially.

I have no idea what this means, but I’m pretty sure that it’s wrong.

Transformation will detrend the data. We will be able to apply statistical models if the data don’t have any trend, in other words predictive models work on white noise.

Why LN transformation? Additivity

Taking a log will not “detrend” the series. Detrending the series would be done by partialling out the component of the DV related to time or some other way accounting for time component of the trend. Your claim that “we will be able to apply statistical models if the data don’t have trend…” is ambiguous and seems to imply data with trend cannot be modeled which is unequivocally false. Your next claim is “predictive models work on white noise” which again is so incredibly incorrect because there are plenty of models that work on non-independent data.

It concerns me that you’ve used so many words and ideas incorrectly and in a manner that might make people believe you know how to use them (and you said it so matter-of-factly).

You generalized the context and not referred the specificity of the problem.From my memory, I think the problem of the inst. was not a complex one where the objective was to apply Linear Regression on an econometrics dataset which follows an exponential trend. I don’t remember anywhere CFA Inst assumed that price data do not follow a LogNormal trend.

I said “Transformation” did not say “LN Transformation” will detrend the data. I am into stats modeling for sometime and I am aware of what I said. In CFA context for example,I will ALWAYS assume derivatives pricing can be done only by assuming Risk Neutrality outside I can think about Risk Sensitive pricing. We need to remember we are studying for CFA not for Financial Engineering

You generalized the context and not referred the specificity of the problem.From my memory, I think the problem of the inst. was not a complex one where the objective was to apply Linear Regression on an econometrics dataset which follows an exponential trend. I don’t remember anywhere CFA Inst assumed that price data do not follow a LogNormal trend.

The CFA institute doesn’t determine what’s good statistical practice or follows with statistical theory. They’re quite far from *any* authority on statistics (or econometrics if you’re so inclinded ot claim that’s very different in this context).

I said “Transformation” did not say “LN Transformation” will detrend the data. I am into stats modeling for sometime and I am aware of what I said. We need to remember we are studying for CFA not for Financial Engineering

I understand what you said, but the OP was asking about logarithmic transforms (a specific detail, but then you claim I made my response too general when you’ve talked about “transformation” in general rather than the OPs specific question). Partialling out is generally not considered a transformation otherwise every linear regression *necessarily* would transform the DV because the right-hand side variables are partialled out of the DV; clearly this isn’t true.

You may work with statistical models, but what you’ve said is at best very unclear (giving an extreme benefit of the large doubt created by your post) and most probably incorrect (as evidenced by what you said). Unfortunately using statistics doesn’t make one knowledgeable or an expert; this is why tons of physicians, psychologists, engineers, epidemiologists use statistics on a daily basis and publish papers, but are a tremendous source of poor-quality, irreproducible research that you might see touted in the news as part of some “research crisis”.

The context of the CFA exams (or financial engineering, which is neither statistics nor would I generally trust a financial engineer with statistical problems) doesn’t suddenly make well-known statistical methodologies or theory incorrect.

There was a lot said in your original post which is easily shown to be both unclear and incorrect. I’m going to leave this here, for I have seen Dunning and Kruger are already present!

Cheers!

I won’t deny the fact that I am pretty new to the industry, hence I need to learn a lot and I should have been clearer. And I never claimed I am an expert while I am not, but I owned what I had written.

I don’t think any Dunning and Krugger effect is present. Rather, I always knew I was talking to an expert campaigner of AF, whose posts in Quants cleared many doubts of myself when I was a candidate.

Peace Sir. Looking forward to learn from you

I won’t deny the fact that I am pretty new to the industry, hence I need to learn a lot and I should have been clearer.

I’m not in the industry :). If you’re interested, I can recommend a good book written by statisticians with some practical examples in finance and they use R. It also comes with video lectures and is free in PDF and only about 40 USD new on amazon. It’s called Introduction to Statistical Learning by Tibshirani, and Hastie.

Peace Sir. Looking forward to learn from you

I am glad I used to be helpful. I apologize if I was rude earlier-- was posting after a bit of work :|… Keep the interest in learning

Tickersu, thanks for the book rec. Always can get better/learn new things as we go. Since we are on the topic, do you have any go to reference sites/books for working on charting skills?

thanks again!

It’s called Introduction to Statistical Learning by Tibshirani, and Hastie.

Thank you much for your reference Sir. I think I am also using the same book, quite popular

tickersu:It’s called Introduction to Statistical Learning by Tibshirani, and Hastie.

Thank you much for your reference Sir. I think I am also using the same book, quite popular

They certainly marketed appropriately for the “Data Science” nonsense out there. Probably the best book for this high-level intro because they actually know what they’re talking about (uncommon in these R and python books, python is far worse). Get to their free course through Stanford with the videos if you can and do the problems in the book so you can graduate to a deeper treatmet (believe it or not, that book is to full treatment of the topics as an itinerary is compared to a full detailing of the days events with video playback).

Thanks again for your guidance