I’ve started taking a couple of the QM subject tests from CFAI, and I had a question for the experienced test takers. First, let me explain what I’m referencing.
In the Hamilton case, I came across a question regarding the Durbin-Watson test, where they gave you only the DW statistic (0.84, was the number). They asked you to make a conclusion regarding the presence and kind of serial correlation. Naturally, I expected the relevant excerpt from a Durbin-Watson table to accompany the question (or even the approximate p-value), but nothing was included. I understand the range of the DW statistic and the implications of values “close” to 0, 2, and 4, but how “close” is usually aided by a table with critical values, or the question is blatantly obvious (because they give you a DW statistic of 0, 2, or 4).
So, will the test be more complete in the provided info than this?
Not to be rude, but I’m not sure you understood what I’m asking. I’m talking about a time series topic, not ANOVA (which CFAI doesn’t truly cover-- look up ANOVA, if you don’t believe me, it’s a technique used to determine differences in at least three means across different (groups) populations). Again, my main concern is that they didn’t provide any crtitical values or approximate p-values to aid in the decision making. In my opinion, it was a poorly implemented question.
I appreciate your reply, but I’m still looking for clarification on this.
whatever that table on page 389 (exhibit 1) is called. for that example there are tables on the following page, however I think it’s expected you know what values are roughly significant.
Exhibit 1 on 389 is a summary regression output and exhibit 2 on page 390 is an excerpt from a 5% significance Durbin-Watson table (this is what you need). This set up is how the problem should be arranged. The practice test (online) I’m referring to doesn’t include any table similar to Exhibit 2 on page 390. All they give you in the online problem is a calculated DW statistic (no critical values). The Durbin Watson critical values vary with sample size, the number of non-intercept parameters estimated, and the significance level, so I can’t imagine that we’re expected to memorize any set of critical values for the DW tests. Sure, I can tell you that it is “roughly” significant, but they’re asking for a clean decision (it is significant or it isn’t).
For example, if I give you a calculated DW statistic of 0.505 and tell you it was for a regression with 3 parameters excluding the intercept and 13 observations (which is not uncommon in time series to have few observations), could you tell me (off the top of your head) if there is positive serial correlation? My guess is that you would say yes, but at the 1% level, you would be incorrect. It would fall in the inconclusive range, which is another reason why I think the idea of “knowing what’s roughly significant” is a useless approach for the DW test (unless you get a value very close to 0,2, or 4 (say 0.15, 1.98, or 3.98, respectively). Then it’s a more fair question.
Take a look at the tables here, and maybe you’ll understand. It’s a lot different than saying someone should know absolute critical z values of |1.645|, |1.96|, and |2.58|. It’s equivalent to saying you should know critical F-values.
i just did that vignette and got 100%. I think the DW question is fair. I have in my mind values between 1.4-1.7 as being ‘questionable’ (and I don’t expect a 2-4 figure, but I could probably figure it out).
if you are at the stage that you are questioning the accuracy of the syllabus then it’s probably time to focus on your weaker topics.(likelyhood is there will only be 1 quant vignette on the exam).
When you reference your 1.4-1.7 range and say if they gave you 2-4 you could probably figure it out, all you’re doing there is guessing (as compared to actually using a DW table or approximate p-value to make the call). Again, in the example I gave about a sample of 13 and 3 estimated coefficients, I’m willing to bet most people would claim positive serial correlation when looking only at the calculated DW statistic of 0.505, but they wouldn’t be correct because the bounds change depending on sample size, estimated coefficients other than the intercept, and the significance level.
I did the vignette and achieved 100% on my first attempt as well. I entirely think it is a fair question (just a hypothesis test, after all), but I think it is a poorly executed question (all they required was a parenthetical remark about the lower and upper DW critical values). I had an idea of what they were trying to test, though, which enabled me to answer correctly.
I’m not sure what you mean by “questioning the accuracy of the syllabus”, because I’m just asking if there will be a greater level of concreteness on the actual exam.