The 2/3 approach for CFA level 1

T

turbo_agni

New member
Joined
Jun 12, 2014
Messages
0
Reaction score
0
I followed this strategy and had about 40 days (summer vacation) and no backround other than quant.
Since there is no negative marking in the exam and only 3 options, the expected score a totally unprepared candidate, marking random choices, would score is 33%.
Given that you are sure of half the questions on the CFA exam and mark the remaining randomly you are expected to score around 50%+0.33*50%=66% which was probably the MPS this year.
So to confidently score 50% I picked and covered 67% of the curriculum (FRA, Book 4. Fixed Income, equity, Quant) and I left out 1/3rd of the topics completely.
I took a mock the day b4 the exam and scored ~56% attempting the 67% I had covered. I finally passed the exam with >70% in all except book 1 and book 6 topics.
What are your thoughts on this approach?? Has someone else passed the exam with a 67% prep.
PS - I’m definitely not encouraging this approach.
 
turbo_agni wrote:
I followed this strategy and had about 40 days (summer vacation) and no backround other than quant.
Since there is no negative marking in the exam and only 3 options, the expected score a totally unprepared candidate, marking random choices, would score is 33%.
Given that you are sure of half the questions on the CFA exam and mark the remaining randomly you are expected to score around 50%+0.33*50%=66% which was probably the MPS this year.
So to confidently score 50% I picked and covered 67% of the curriculum (FRA, Book 4. Fixed Income, equity, Quant) and I left out 1/3rd of the topics completely.
I took a mock the day b4 the exam and scored ~56% attempting the 67% I had covered. I finally passed the exam with >70% in all except book 1 and book 6 topics.
What are your thoughts on this approach?? Has someone else passed the exam with a 67% prep.
PS - I’m definitely not encouraging this approach.
Click to expand...
I was thinking about a similar approach, though perhaps skipping a total of 10% of material, not 33%! I’m about to start book V of Kaplan (derivatives and alt. investments) but will likely skip a few small sections when I do my final review (going over Kaplan again while devoting a good portion of time to CFAI).
I know some people who only did a very high level of review for Econ (10 hours at best) then didn’t do any other review. Econ constitutes the largest block of material yet only represents (10%?) of material on the exam. It makes sense - Econ is a drag.
 
I punted on Econ at L1 - too little reward for so much reading. It showed in my score matrix, but I never feared failure.
 
don’t these topics build on each other in future levels so you might be screwed going forward
 
Uhh risky.
I almost skipped Econ. I mean I was trying to learn it but I was already totally burned out so nothing stuck in my head, but otherwise I covered everything. I really thought you cannot “officially” pass if you have anything <50%.
Well I passed with Econ <50 (FI+Ethics < 70 and rest above 70).
I sometimes think I shouldn’t have bothered with Econ at all, and instead (1) have some rest, (2) do some more mocks.
To skip 1/3 is maybe too much, but otherwise I support the strategy of knowing part of the subjects almost 100% (and part of them much less) than aiming for a general “know a bit of everything” method.
 
robustening wrote:
don’t these topics build on each other in future levels so you might be screwed going forward
Click to expand...
FRA yes, rest I don’t know. Econ surely not. (Hopefuly).
But I think this strategy is if you are under time pressure and you have to take a risk anyways.
 
Congratz on the pass, amazing work. But I don’t think I’ll be taking this approach
turbo_agni wrote:
Given that you are sure of half the questions on the CFA exam and mark the remaining randomly you are expected to score around 50%+0.33*50%=66% which was probably the MPS this year.
So to confidently score 50% I picked and covered 67% of the curriculum (FRA, Book 4. Fixed Income, equity, Quant) and I left out 1/3rd of the topics completely.
Click to expand...
The logic is a little off - two problems. You are first assuming getting 50% of the whole exam, including the unstudied part; and you assume to get 33% right, including your incorrect answers on the studied part.
You need to weight the probabilities:
weight of studied material * percent correct + weight of unstudied material * percent correct = 0.66
0.66 * x + 0.34 * 1/3 = 0.66
x = 0.828
You have to guaruntee getting 83% or higher on the material you do study.
Also, if you answer a question completely randomly then 33% is the expected mark, but if you read the question and give it your best guess all estimates are off. Maybe you get some right, maybe you fall into some ‘trap’ answers. Who knows. You can’t expect to get 33% correct.
 
oktavian wrote:
The logic is a little off - two problems. You are first assuming getting 50% of the whole exam, including the unstudied part; and you assume to get 33% right, including your incorrect answers on the studied part.
Click to expand...
I meant it a bit differently. You get your paper. U are ultra-confident about half the questions and randomly guess the remaining. That way you have -
1*0.5+.33*.5 = 0.66
So what I meant was, to know 50% of the paper with certainity, you might consider properly studying 2/3 of the material.
Yes I do agree that you need to know the material well enough to score above 83%.
And only follow this approach if you have a serious time constraint!!
 
Yes I agree. I just looked at the CFA L2 material I don’t think there’s Econ in it at all. And Alternate investment and derivatives were not covered in depth in L1 any way and the basics you’d need are well laid out in L2.
True you’ll have to go through Ethics
 
Google CFA Level I practice exams and take them and see if you score 30% with no study. You might find your assumptions have to be adjusted. Know Ethics cold or you will fail.
 
robustening wrote:
don’t these topics build on each other in future levels so you might be screwed going forward
Click to expand...
Some do and some don’t.
Econ really doesn’t. Econ at the higher levels is independent of L1 stuff. Even then, it’s a very small portion of L2 and 3.
Statistics definitely builds. But again, it’s such a small portion of L2 and 3 that you might be able to punt it. (I’m talking about the statistics portion of quant. You definitely need to understand the TVM stuff.)
The other stuff at L1 all definitely builds. EG - You can’t just ignore the options stuff, because you will be completely lost at L2 and 3. You have to have a good base of knowledge on everything on L1 if you expect to pass the higher levels.
 
Try the “1” approach or just go home… this test isn’t one you want to try to game
 
Justbeat wrote:
Google CFA Level I practice exams and take them and see if you score 30% with no study. You might find your assumptions have to be adjusted. Know Ethics cold or you will fail.
Click to expand...
It’s basic statistics that the expected score of a totally unprepared candidate is 33%. Just to convince you I tried out what you said. I’ve written a small MATLAB code to do the job for me and guess what?? The score followed a normal distribution with a mean of 33.25% and a standard deviation of 4.1427%. U can find the code below -
function [Mu,Std_dev] = CFA_L1()
Correct = zeros(30,1);
for n = 1:30
Actual_Ans = floor((4-1)*rand(120,1)+1);
Answers = floor((4-1)*rand(120,1)+1);
Correct(n) = (1-nnz(Actual_Ans-Answers)/120)*100;
end
Mu = mean(Correct);
Std_dev = sqrt(var(Correct));
end
I do agree that to be taking a less of a risk, you would want to lie ahead of Mu+2*sigma. But taking into account that a candidate can in many questions eliminate atleast one option kicks up his probability to atleast 33%.
Though I’m not for this approach I feel that a candidate who has throughly read through 2/3 of the curriculum and has a superficial knowledge of the remaining, he/she can peacefully pass the L1 exam.
No comments on this approach’s effect on L2 & L3. Just speaking in terms of numbers.
 
IsThereAny wrote:
Try the “1” approach or just go home… this test isn’t one you want to try to game
Click to expand...
Hey!! I’m not trying to game the system here.
I’ve lately been reading many posts of people fretting over not remembering everything. I just feel that, it’s atleast not as tough as it looks, and it’s OK to not remember somethings before the exam because it statistically looks very probable that you could pass with a good 2/3 prep and a somewhat superficial 1/3 prep.
 
turbo_agni wrote:
Justbeat wrote:
Google CFA Level I practice exams and take them and see if you score 30% with no study. You might find your assumptions have to be adjusted. Know Ethics cold or you will fail.
Click to expand...
It’s basic statistics that the expected score of a totally unprepared candidate is 33%. Just to convince you I tried out what you said. I’ve written a small MATLAB code to do the job for me and guess what?? The score followed a normal distribution with a mean of 33.25% and a standard deviation of 4.1427%. U can find the code below -
function [Mu,Std_dev] = CFA_L1()
Correct = zeros(30,1);
for n = 1:30
Actual_Ans = floor((4-1)*rand(120,1)+1);
Answers = floor((4-1)*rand(120,1)+1);
Correct(n) = (1-nnz(Actual_Ans-Answers)/120)*100;
end
Mu = mean(Correct);
Std_dev = sqrt(var(Correct));
end
I do agree that to be taking a less of a risk, you would want to lie ahead of Mu+2*sigma. But taking into account that a candidate can in many questions eliminate atleast one option kicks up his probability to atleast 33%.
Though I’m not for this approach I feel that a candidate who has throughly read through 2/3 of the curriculum and has a superficial knowledge of the remaining, he/she can peacefully pass the L1 exam.
No comments on this approach’s effect on L2 & L3. Just speaking in terms of numbers.
Click to expand...
Yep, if you answer completely randomly, you will expect to get 1/3 correct. But humans can’t make random choices. Choosing all ‘a’s isn’t random. Choose ‘a’ then ‘b’ then ‘c’ isn’t random. But the biggest thing, most people will try their best guess to answer correctly (probably the best move..?), and that is not answering randomly. So this figure of 33% is just out the window.
But… you passed, the only statistic that counts..!
 
I am not sure whether this is a tongue-in-cheek or serious contribution.
Whatever might be the case, I’d like to clarify, for the benefit of those who are starting out in statistics, the following:
1) The Binomial distribution converges to the normal as the sample size (n) goes to infinity. In practice good approximations can be achieved for finite, even small n, but the above statement “The score followed a normal distribution” is factually incorrect.
2) There is no point in referring to the normal distribution, as the given application’s natural distribuation is the Binominal; as the latter can be calculated most easily in closed form, switching to the normal introduces unnecessary error. This is especially true for events closer to the tails, such as “two sigma” events, as even a slight degree of skewness will have a material impact on the prob.
3) The standard deviation of a Binomial random variate is sqrt(n*p*(1-p)). Plug in the figures and divide by the sample size n to arrive at the accurate “percentage” stdev quoted in the post to be 4.1427% (BTW no serious scientiest or statistician would present more than three significant figures, especially if, as is the case here, the figure is inaccurate => see point 4)).
4) If one needs to employ Monte Carlo to integrate numerically, then a sample size of one, as used here, is not sufficient. The whole point of MC is to arrive at an accurate expectations by simulating the complex more than once. The resultant distribution allows one to give accurate estimates of the required parameter (here mean and stdev of Binomial distriburion).
I personally feel that it is unfair to confuse forum Level 1 participants, who might not have had prior exposure to basic statistics, with such misleading and partially also factually simply incorrect contributions.
So I hope this was some sort of practical joke?
 
Dear tpmk,
tpmk wrote:
The Binomial distribution converges to the normal as the sample size (n) goes to infinity. In practice good approximations can be achieved for finite, even small n, but the above statement “The score followed a normal distribution” is factually incorrect.
Click to expand...
Each individual question is a random variable with an outcome of either 1 (correct) or 0 (wrong) with probabilities as 1/3 or 2/3 respectively. Clearly this random variable follows a binomial distribution.
Now since all the questions (random variables) are independent and identically distributed, N = 120 of them can be combined to form a sample. Now from CENTRAL LIMIT THEOREM, the arithmetic mean (%correctly answered) of the sample follows a Normal distribution, especially when ‘n’ is large (30 in my case).
So please get your concepts right before you label something as “factually incorrect”
tpmk wrote:
The standard deviation of a Binomial random variate is sqrt(n*p*(1-p)). Plug in the figures and divide by the sample size n to arrive at the accurate “percentage” stdev quoted in the post to be 4.1427% (BTW no serious scientiest or statistician would present more than three significant figures, especially if, as is the case here, the figure is inaccurate => see point 4)).
Click to expand...
Now that we are talking about the “Mean of the sample” as the random variable, your formula no longer holds true. Refer to http://en.wikipedia.org/wiki/Central_limit_theorem for the correct formula.
Thank you correcting my mistake. I copy pasted the stdev from the command window. FYI I am no serious scientiest or statistician and I’m sure you are not one either.
tpmk wrote:
There is no point in referring to the normal distribution, as the given application’s natural distribuation is the Binominal; as the latter can be calculated most easily in closed form, switching to the normal introduces unnecessary error. This is especially true for events closer to the tails, such as “two sigma” events, as even a slight degree of skewness will have a material impact on the prob.
Click to expand...
What’s your point bro??
tpmk wrote:
If one needs to employ Monte Carlo to integrate numerically, then a sample size of one, as used here, is not sufficient. The whole point of MC is to arrive at an accurate expectations by simulating the complex more than once. The resultant distribution allows one to give accurate estimates of the required parameter (here mean and stdev of Binomial distriburion).
Click to expand...
I think you are trying to shoot a mosquito with a revolver here!! It’s a simple scenario. There’s no need for MC.
So it’s you who is confusing forum Level 1 participants, who might not have had prior exposure to basic statistics, with such misleading and partially also factually simply incorrect contributions.
ALL DE BEST for your endeavours in life
 
You might not be encouraging this approach, but I think it’s still inspriing/inspired stuffed. Well done and thanks for sharing.
I am in a similar situation, i.e. have got a background in quantitative methods and will have to make do with a couple of months worth of preparation.
Wish me luck!
 
All the best buddy!!
It’s not as difficult as it seems you know. Please don’t frett over not remembering every detail of the curricullum.
That’s what I wanted to highlight in the post - you’ll mostly pass with a good 2/3 prep and a superficial 1/3 prep.
 
You must log in or register to reply here.
Back
Top