Because I haven’t got a clue what the hell the Time Series study session is about. Multiple Regression is quite easy but then I read that sh*t and I don’t even know what they want me to understand from it. Is it just an elaboration of the previous study session and its heteroskedasticity?
I have seen this reaction on other threads too, as well as with me. I had to read it three times before I grasped the material. It’ll come to you. Just re-read it.
You are not alone- believe me. I am taking Level II again this year. Much of what is review from last year seems easy. But the new stuff seems really difficult- like time series. I don’t think it really has all that much to do with intelligence. Some people have finance backgrounds and it is an easier learning curve, for some the curve is steeper- especially for us English majors. I don’t for a minute believe not understanding the first time is any reflection of how smart you are. You could make an arguement that the really intelligent people don’t need to take these tests anyway-
I got the time series pretty well first time through, the multiple regression is what I had to go back through and re-read. So who knows with this stuff, everyone keeps telling me multiple regression wasn’t that bad which I guess puts me in the “special needs” category.
Can anyone give me with a real world example of what a “unit root” is. Every explanation I read just makes me more confused.
I could use a good explanation on that too. Joey?
There is no easy way to explain unit root. It has to do with solving an equation and looking at an intersection of the unit circle. The typical unit root equation is something that’s autoregression order 1 such as yt such that dyt=e (dyt=yt-yt-1 and e=error). This means that yt is non-stationary. You can also say that it is integrated of order 1. This is bad when you run a regression because if the two variables you run a regression against both have unit roots then you actually have what’s called a spurious regression. The relationship only appears to exist because they both are actually generated from the same process. There is no causal relationship. The solution is described in the CFAI notes. You have to take the first difference (or ln difference, mostly used in finance) if you’re just working with one unit root. If you’re working with two, you can use Engle-Granger. Basically regression them against each other, take the residuals and if both of them together have no unit root, they are cointegrated. Ultimately, you can use this to create an Error-Correction Model. I studied these topics in a Graduate level Time Series class and I still don’t understand them very well (they get much more complex). I do know that the CFAI books explain the topics much better than my teacher ever did. So I would reread those sections. If you have any other questions, feel free to send me a message.
Does anyone else feel like they are forgetting things every time you move on to another book? I felt like I had a good handle on the multiple regression / time series material when I was on the quant / econ book. Now, not so much. I think Quant and FSA are going to get some extra attention once I finish the rest of the curriculum…
I’m glad I’m not the only one! This motivates me to give it another try today. Must… understand… time series…ugh
Yeah, alot of the areas I’ve studied are not firm in my mind at all, so I think that by continualy answereing Qbank and the my end of study review & Secret Source , should solidify it all.
Can someone explain the difference between a Random Walk and Coveriance Stationarity? I don’t get it.
A random walk simply means Yt=a*Yt-1+e and a=1. In other words the change in a variable is random. Covariance stationarity, I prefer to just say stationarity, is more general than a random walk. It means that something has the same properties over time and does not change (we call them moments in statistics: mean, variance, skewness, and kurtosis are the first four). Strictly speaking, covariance stationarity just means that the process just has a mean and variance the same over all time periods. If something has a unit root, as noted above, it is not stationary. A random walk is not stationary since it has a variance that depends on the time period. As you continue randomly walking away, the spread will get wider and wider, and variance gets larger. You can prove that something has a unit root with the ADF test and generally if you show that it is not integrated order 1 or 2, then you can assume that it is stationary (though that really isn’t scientifically proper). If the process does not depend on past elements of itself, such as yt=e, then it makes sense to say that it is stationary.
In the “don’t feel stupid” category, the Stalla Instructor from the videos said that this is a topic that won the Nobel Prize in Economics a few years back, so it certianly is not easy. He said don’t get too bogged down, and to focus on the application/how anaylsts use it rather than how the theory/calculations.
Yea, Engle won it just a few years ago (though the idea has been around longer than that). He also won it for ARCh and GARCH. It is important to know about if you run a lot of regressions and so that you don’t make mistakes when you’re doing regressions. It is mostly used in finance for spread trading and other serious quant stuff. Esp. cointegration.
I feel stupid too when I have difficulties grasping seemingly simple stuff; the worst part is that it is dawning upon me that I’ll have to be able to actually do calculations on not only time series but the entire curriculum. How am I to remember all of this by myself, in June, and no cheat sheet either, just the HP12C for comfort… though it’s good to know that we’re having not only a pressed 1.5 minutes per question but the entire eternity of 3 minutes per Q. *grin*
Although the material is definitely hard to grasp, I havent seen questions that really require a full understanding. They always provide you numbers from the ANOVA table, which eliminates the need for serious calculations. It feels like they give you all these hard concepts, but can’t really test you on it.Anyone else get this feeling?
Yeah, the 2006 LII exam questions at the back of some of the readings seemed much more basic than the calculated non-multiple choice questions in other readings. The 3 minutes a Q was bad news for me, I liked 1.5 because it limited the depth of calculations they could expect.
I feel stupid because I have just began studying for L2 last week…I dunno…I guess its kinda stressful to be looking for a job and studying for Level II at the same time…