# exhausted..giving up the crazy Time Series

I spent around one day trying to understand what it say…But totally got lost. This is topic is new in 2008 so I think it’s highly possible to see in exam. But anyway, can’t understand what it mean… autoregressive conditional heteroskedasticity…drive me crazy!!! anyone has idea to deal with this part?

I am with you on this…I am almost sure it will show up on the exam too but cannot get it…I am going to have to read the dreaded CFAI 50 pages on it…

it should be on the exam but i doubt it will be tested heavily… maybe 2 out of 120 questions… so weight your study accordingly

I’m flipping through schweser notes, then solving their questions, qbank questions and the related CFA questions too. Hope to bang enough questions and do enuf corrections in record time to understand it…

I had the same problem with the time series, but somehow managed to understand little bit. This is what I got. Not sure though. Linear - amount growth Log linear - rate growth Plot data on the x-y axis to see covariance stationary - if covariance not stationary with linear, use log linear. Test for serial correlation using DW (only for linear and log linear) If serial correlation exits with linear or log linear, we should not be using either. Use AR model. AR model - Serail correlation test using auto correlation, same as t-test. Increase AR 1, 2, 3 until we resolve serial correlation. Example inclde seasonality, add one more lag representing that season. After that, test for covariance stationary problems such as unit root, random walk. Unit root - test with transformed version and DF, H0: g1=0. First diffference is the solution for unit root. Random walk, I have difficulty in understanding but tried… Randon walk. symptoms are: with drift: b0 not equal to zero and b1=1, first difference is the solution without drift symptoms: b0=0, b1=1. First difference is the solution. ARCH - same test as it is in multiple regression. Regress squared residuals. If a1 (same like b1) is not equal to zero, ARCH is present. Solution ARCH 1 or GARCH models. Unit root in two series - Use two series if both don’t have unit root. Use them if both have unit root and ther are co-integrated. In all other situations, don’t use those series. I don’t think CFAI wants us to be a quatitattive expert. They just want us to understand the concept so that we can detect problems (if any) when we get regerssion results. In the exam, they give regression result and possible problems ( I guess). We just need to suggest which test to use and suggest some solutions. Hope it helps and if I were wrong, somebody help us.

“it should be on the exam but i doubt it will be tested heavily… maybe 2 out of 120 questions… so weight your study accordingly” One thing is for sure; you dont understand how Level II works in comparison to Level I.

what do you mean, caspian??

he means that there can be an entire vignette on it.

ah… like TB last year…

famous last words, but i’ve found the harder the subject, the easier the questions i.e. what type of test should you use?. there have been a few exceptions of course. and i have to agree on the time math too. going crazy for one or two Q’s isn’t worth it (although i fall into same trap myself).

“what do you mean, caspian??” “he means that there can be an entire vignette on it.” Skip a topic and when it gets tested, you’re dead in the water. Level I, you can get by skipping something. But on Level II, it doesnt work like that. If you assume you need 67% of questions right to pass the exam (40 out of 60 on each part), if you miss an entire vignette, then you need 75%-ish of the remaining questions (on each part) to just get the same score.

Anyway, you spent 1 day trying to learn time series analysis? A whole day? What makes you think that learning something like that should be quick?

I am finding Time Series really tough - not sure if it’s a valuable use of time at this stage to try and figure it out. CFAI text looks even more tricky than Schweser to figure out and double the number of pages to read…

let me rephrase that - it would probably be a very valuable use of my time to learn time series material given the high probability it is on exam, but it is not making any sense to me. think I’m going to just skim read the chapter and try to get 2 or 3/6 on a full vignette when it comes up next sat and hope to crush the other QM vignette…

I wouldn’t to go as far to say the Time Series Analysis is worthless. But its certainly worthless for me.