Exam Feedback November 2017 Part 2 Exam Feedback

sergiogh7

New Member
Also, there was a question about LTCM and 2007 subprime credit crisis. I think that the response is that in both cases the problem was they didn’t take properly the correlations in their models.
 

frankdc11

New Member
Does anyone else remember a question which mentioned back testing with a one tailed test?
This is a question I went back to... You were back testing one day 99% VAR and realized exceptions 2% of the time. I went with the logic that you would expect exceptions just 1% of the time for the model to be correctly calibrated. So I narrowed it to the two answers that said it was incorrectly calibrated... But The wording was confusing in the rest of the answers and I think I chose something like "Incorrectly calibrated because you would expect 2% exceptions less than 1% of the time."

if I'm remembering the response correctly, I believe it was option B. I could be completely wrong in my thinking. For some reason this question confused me completely the first time around, so I marked it to go back.
 

yeoshuiming

New Member
Hi, once again, I would like to thank Bionic Turtle for their great notes and practise questions! After doing the practise sets, global drills (credit risk global drill was fun) it probably covered more than the actual GARP test can cover! Yes, its great that now we are only allowed to share feedbacks 1 day after so its fair to everyone regardless of who's taking earlier or later!
What i can recall after 1 day of lag :(( Mine was the orange paper, not blue
Disclaimer: Most details below could be inaccurate. DYODD.

1. TRS to hedge away both credit and market risk. TRS buyer eqv. to funded CLN seller
2. LIBOR forward was 5% since y1=4%, y2=4.5%... an estimate, D.8% was too far off, A,B was below 4% so doesn't make sense
3. why banks not willing to adopt bitcoins? I think its that clearing/validation is done by miners, and because its decentralised and anonymous, banks wouldn't want to shoot their own foot and get involved in future litigation issues by funding potential terrorism...
4. NRP outcome was that banks did not passed on NR to retail depositers, shielding them so as to prevent bank runs when insured, stable retail deposits have a higher run-off to seek higher yields like insti does in riskier investment alternatives...
5. KMV is empirical with a direct default threshold so problem faced could be lack of
6. HFT intraday risk cannot be done by internal team like you will not make an accountant a cashier... so let an independent risk team to monitor and set intraday risk limits
7. RAROC with qns on changing both EC and ROA testing on ratio consistency. so the answer that is not ambiguous is the option that's ratio inconsistent... EC is in both numerator and denominator
8. stats arb hedge fund worst case scenario would be its funding cost besides unprepared for blowouts
9. distressed hedge fund worst case scenario would be an increase in default correlation among distressed securities, given that their strategy is investing in gone concern securities with much higher yield, betting on the unlikely scenario that they thought otherwise
10. CDS-Bond basis is not due to higher funding cost as bond yields will get higher (since bonds are funded) but bond being priced over par
11. NSFR deposits and loans weight need to be clear before you can balance the equation given making 5m loans to auto and the right funding option was clear
12. For rebalancing portfolios, stratified screening is not free from bias was correct because the rest were wrong?
13. CCS = IRS (fixed) payer + long FX fwd, all of similar maturities
14. Outsourcing HR duties i think independent background checks is more important than the rest since HR is not a model so no priority for integration?
15. Risk mapping for bonds from principal to cashflow necessary means a lower risk measurement because of weaker sensitivity (lower duration) for fledging maturities, even if correlation between KR01(i) are perfectly 1 such that KR01 is addictive... aka principal to duration (need not be cash flow)
16. Ops risk: I think external audit of risk process was 3rd line of defence
17. uptake of external debt from emerging economies increases when global liquidity is high is the correct ans?
18. external debt is free from domestic market illiquidity risk is correct rather than trading corporate facing more fx risk than non-trading corporate since I believe FX derivatives are typically used to hedge translation risk?
19. Change from BI to SMA... SMA can charge a higher ops capital charge than BI if there is significant correlation? Tricky and unlikely ans which i think is correct?
20. Model risk: actual culprit in both 2008 Submprime and LTCM cases was because of overly optimistic forecast aka lower than actual default correlation
21. risky debt value = risk free debt value - put option premium... NOT call option for equity valuation
22. long term assets funded by short term debt. Which funding liquidity risk? Rollover of short term debts
23. Rise in DVA caused when its own credit spread decrease along with an upgrade of its posted collateral quality
24. Credit VaR, fortunately given the corresponding 95% number of defaults which is 5, and expected is 100 x 2% = 2, so LGD*notional*(5-2)
25. Netting diversification benefit for a clearing member when it change from bilateral to CCP netting... I think it will not be effective as the MTM was negative
27. Incremental CVA between 1-4 [1 = IRS payer, 4 = CCBS] favors execution sequence of the pair with the most inverse correlation to minimise CVA charge from cpty
28. Regulatory arbitrage between loan book and trading book... No difference under Basel III IRC and that its CDS that hedge away its credit risk it not recognised under trading book
29. 3x qns on correcting infrequent funds performance reporting with autocorrelation - one on mean reversion which is 1-autocorr, another is tricky... returns are unaffected and, lastly also another tricky one... (where managers arguing for unsmoothing of returns) PM's sharpe ratio increases as volatility falls
30. Electronification of fixed income market reduce dependence of dealers to provide liquidity
31. Fama 3 factor model analysis on a fund's performance where negative coefficient on HML means its portfolio low books value stocks are outperforming its high book value stocks
32. A qns on right way risk... where A long put options on Z and B is cpty to A who is short Z.. so what happens when Z price falls, A experience Right Way risk
33. A qns on Advanced IRB which uses a single factor (market) model and a multivariate coupla correlation input... credit risk capital charge increases when correlation or standard deviation (any 2nd moment variables) increases which increases the tail risk
34. Stressed Loss for credit "derivatives" in loan book need both stressed PD and stressed EPE (since derivatives are "uncollaterised, leveraged exposures" and more sensitive to credit deterioration risk leading to higher PD and default risk leading to EAD) and not stressed current exposure to fix EAD to deterministic, with k multiplier and EL given
35. RAF (Risk Appetite Framework) on business lines to be set by self assessments or independent analysis of that business line loss distribution from both internal and external data
36. LIBOR (unsecured credit lines to AAA cpty) - OIS (can be terminated and reset easily without excessive credit risk) rather than LIBOR-FF (only for assets eligible to discount window) or FF-OIS as a monitor of potential funding liquidity risk
37. When yield curve steepens, duration of debt increases resulting in higher credit risk on liabilities side of the balance sheet of a fund issuing debt to fund its asset acquisition... so if curve steepens after acquisition of new assets by issuing more debt, and fund change from principle to duration risk mapping, risk budget should increase
38. qns on ISDA where its on break clauses as triggers to give bond holders the rights to terminate swaps before cpty risk migrates lower, but its weakness is the definition of a credit event might only materialise after news that point towards a credit deterioration being very likely
39. Best practise of risk budgeting? Should stress those factors which business lines are most sensitive to?
40. mVaR ~ Beta and excess return was given, but qns ask for optimal "ratio of alpha to portfolio risk contribution" so I trim lowest excess return / beta, vice versa for adding
41. Why did 97.5% ES == 99% VaR when it started adoption of SMA for OpsRisk? I think he mistaken a normal distribution for severity of loss... loss should be skwed with a long tail?
42. issuance of convertible bonds (CoCos) impact on balance sheet and ROE?
43. Backtesting of VaR estimates is tricky since it should converge to 1 default not 2 exception failure rate as its a 99% hypothesis test on 99%,10-days VaR. So the better of 2 angels (incorrectly calibrated) was the answer?
44. Vasicek was tricky as you need to update the MR effect on rate change step-by-step for 4x 6m steps for a 2y period, but difference is negligible
45. Copulas curse of dimensionality rule most options out, leaving the brightest angle as accepting non-normal distributed inputs
46. Replicating portfolio with long beta amount of single factor market exposure with the rest in risk free debt?

Some easier qns (most starting from qns 50 onwards were manageable):
- asset allocation returns
- marginal PD given cumulative #defaults
- netting and collateral balance
- CDS spread / (1-RR) = PD, bond yield was irrelevant?
- credit spread given discount factor and risk free rate
- exogenous LC
- regression single hedge
- expected lee-ho model given annualised drift and rate volatility but no dw
- calculation of incremental VaR from 1 asset to 2 assets, which is usually done with the help of computers
smile.png
so quite arghh...

General comments:
- most questions should not need a calculator, and could be approximated base on concept... (I could be wrong big time here... correct me if I'm wrong) E.g. Liquidity Cost cannot be >70% of VAR w/o LC (exo), incremental VaR with -0.2 corr cannot have >= individual, undiversified VaR... calculators are meant to burn away your precious time
- many option set questions where its always a better of 2 "angels" decision for qns that tunes with "choose the correct option"...
- Some questions were "resused" from May 2017 FRM Part II.. so I'm expecting the distribution to have a thicker right tail. E.g. vol term structure was a frown for expected bimodal return distribution in a binary event favoring strangle, LIBOR discounting to determine LIBOR forward rate, Bitcoins, NRP, marginal PD, wrong way risk has the highest CVA (producer NOT hedging), CoCos...
- I expect the quantiles to be right skewed this time round... its hard!
Cheers!

J.
18 Nov 2017, GARP FRM Part II
Singapore, Bayfront Expo, Hibiscus Room
 
Last edited:

kik92

Member
This is a question I went back to... You were back testing one day 99% VAR and realized exceptions 2% of the time. I went with the logic that you would expect exceptions just 1% of the time for the model to be correctly calibrated. So I narrowed it to the two answers that said it was incorrectly calibrated... But The wording was confusing in the rest of the answers and I think I chose something like "Incorrectly calibrated because you would expect 2% exceptions less than 1% of the time."

if I'm remembering the response correctly, I believe it was option B. I could be completely wrong in my thinking. For some reason this question confused me completely the first time around, so I marked it to go back.
I cannot remember what answer I chose however the fact that they mentioned a 'one tailed' back test was incorrect am I wrong? shouldn't back testing be two tailed !
 

davidhn

New Member
Hi all,

I spent about an hour trying to remember as many questions as possible yesterday while going through the material - I came up with about 55. I can post them tonight if anyone is interested. After a quick check I made roughly 15 mistakes out of 55, so it seems from this very approximate sampling that I should be a notch above 70%, so I am only mildly confident.

@frankdc11, I answered this question differently but I might be wrong. I am a bit foggy on the details though but I think I remember it indeed said that there were realized exceptions 2% of the time (not entirely sure about that though).

We are backtesting over 250 days with 99% confidence so we expect 0.01*250 = 2.5 exceptions.
We end up with 2% exceptions, or 0.02*250 = 5 exceptions.

(5-2.5)/sqrt(0.01*0.99*250) = 1.589, which is inferior to the critical value needed to reject the hypothesis that the model is correctly calibrated.

In other words, 2% exceptions over 250 days instead of the expected 1% is not statistically significant enough to conclude with 99% confidence that the model is flawed.

Other questions I remember from the top of my head / my answer / how correct I believe my answer is after quickly checking the books (happy if you can prove me wrong either way!) :

1/Bitcoin risk / being used for transactions in sanctioned countries / correct

2/Collateral to be posted and balance / 5 and 7 (5 above threshold + 2 INDEPENDENT amount) / correct

3/Similiarities between 2007-2009 and LTCM / correlation risk understimated / correct

4/Implied volatility shape before announcement / volatility frown / correct

5/Credit spread using -ln(yrs to maturity) * ln (market value) - ln(face value) / can't remember / correct

6/Which instrument to hedge credit risk and market risk / TRS /
unsure but I believe TRS does hedge market risk for the buyer, since seller/receiver has to pay any depreciation in the underlying asset?

7/A question on 99% ES where they gave you 99.2%, 99.4%, etc. values / I averaged starting from 99.2%, since ES should be loss IN EXCESS of the quantile; not 99% / I believe this is
correct

8/A lengthy scenario involving 3 companies with positions with each other in a call, CDS, etc. and possible answers with right way risk - what the hell? I thought this one was a bit absurd. I eliminated "call price would decrease" since the underlying price and volatility rose, which both make a call more expensive, but then I wasn't sure so I picked "increase in right way risk" since all the companies seemed to be improving, but I really had no idea so I am gonna go with
wrong
/
9/A calculation of required market capital / do not forget to add the stressed VaR component and use multiplier /
correct

10/Indicator of decreasing funding liquidity / increasing treasury-corp spreads /
wrong I believe; there was an option with dealer bank clients reducing their exposure which I think was better

11/Value of risky debt under Merton / Risk-free minus put /
correct

12/Incremental VaR with negatively correlated position / the negative answer (position actually reduces ptf VaR due to negative correlation) /
correct

13/LIBOR-OIS discounting / I didn't know how to do that so I went with 5% but I believe it is
incorrect (as I did not even use OIS rates at all and just went for a simple "implied 1 year-rate in 1 year" calculation)

14/Vasicek expected rate after 4 years / current*exp(-k*T) + long.run.value(1- exp(-k*T)) / correct

15/Ho-Lee after 2 periods in lower node / first drift/12 + second drift/12 - 2*annual vol*sqrt(1/12) since lower node is following the "negative shocks" path twice / correct

16/17/2 questions about Fixed Income VaR Mapping / I hate this concept and always had a weak understanding of it so I don't remember / am gonna go with incorrect

18/Which asset to add to another to form a 2-asset portfolio given a risk budget of 67 million if I remember correctly (200 million added to 300 million) / calculated the answer that seemed obvious just by reading it (low correlation with existing asset) / correct

19/Some stuff about best practice when outsourcing / comparing the IT systems or controls or something like that / unsure

20/Bond price to discount using a tree and 1 + r/2 / can't remember / most likely correct
 

bake5472

New Member
Hi all,

I spent about an hour trying to remember as many questions as possible yesterday while going through the material - I came up with about 55. I can post them tonight if anyone is interested. After a quick check I made roughly 15 mistakes out of 55, so it seems from this very approximate sampling that I should be a notch above 70%, so I am only mildly confident.

@frankdc11, I answered this question differently but I might be wrong. I am a bit foggy on the details though but I think I remember it indeed said that there were realized exceptions 2% of the time (not entirely sure about that though).

We are backtesting over 250 days with 99% confidence so we expect 0.01*250 = 2.5 exceptions.
We end up with 2% exceptions, or 0.02*250 = 5 exceptions.

(5-2.5)/sqrt(0.01*0.99*250) = 1.589, which is inferior to the critical value needed to reject the hypothesis that the model is correctly calibrated.

In other words, 2% exceptions over 250 days instead of the expected 1% is not statistically significant enough to conclude with 99% confidence that the model is flawed.

Other questions I remember from the top of my head / my answer / how correct I believe my answer is after quickly checking the books (happy if you can prove me wrong either way!) :

1/Bitcoin risk / being used for transactions in sanctioned countries / correct

2/Collateral to be posted and balance / 5 and 7 (5 above threshold + 2 INDEPENDENT amount) / correct

3/Similiarities between 2007-2009 and LTCM / correlation risk understimated / correct

4/Implied volatility shape before announcement / volatility frown / correct

5/Credit spread using -ln(yrs to maturity) * ln (market value) - ln(face value) / can't remember / correct

6/Which instrument to hedge credit risk and market risk / TRS /
unsure but I believe TRS does hedge market risk for the buyer, since seller/receiver has to pay any depreciation in the underlying asset?

7/A question on 99% ES where they gave you 99.2%, 99.4%, etc. values / I averaged starting from 99.2%, since ES should be loss IN EXCESS of the quantile; not 99% / I believe this is
correct

8/A lengthy scenario involving 3 companies with positions with each other in a call, CDS, etc. and possible answers with right way risk - what the hell? I thought this one was a bit absurd. I eliminated "call price would decrease" since the underlying price and volatility rose, which both make a call more expensive, but then I wasn't sure so I picked "increase in right way risk" since all the companies seemed to be improving, but I really had no idea so I am gonna go with
wrong
/
9/A calculation of required market capital / do not forget to add the stressed VaR component and use multiplier /
correct

10/Indicator of decreasing funding liquidity / increasing treasury-corp spreads /
wrong I believe; there was an option with dealer bank clients reducing their exposure which I think was better

11/Value of risky debt under Merton / Risk-free minus put /
correct

12/Incremental VaR with negatively correlated position / the negative answer (position actually reduces ptf VaR due to negative correlation) /
correct

13/LIBOR-OIS discounting / I didn't know how to do that so I went with 5% but I believe it is
incorrect (as I did not even use OIS rates at all and just went for a simple "implied 1 year-rate in 1 year" calculation)

14/Vasicek expected rate after 4 years / current*exp(-k*T) + long.run.value(1- exp(-k*T)) / correct

15/Ho-Lee after 2 periods in lower node / first drift/12 + second drift/12 - 2*annual vol*sqrt(1/12) since lower node is following the "negative shocks" path twice / correct

16/17/2 questions about Fixed Income VaR Mapping / I hate this concept and always had a weak understanding of it so I don't remember / am gonna go with incorrect

18/Which asset to add to another to form a 2-asset portfolio given a risk budget of 67 million if I remember correctly (200 million added to 300 million) / calculated the answer that seemed obvious just by reading it (low correlation with existing asset) / correct

19/Some stuff about best practice when outsourcing / comparing the IT systems or controls or something like that / unsure

20/Bond price to discount using a tree and 1 + r/2 / can't remember / most likely correct
You said you remember 50+... keep them coming :)
 

frankdc11

New Member
@davidhn I calculated the same 1.589 on my first pass of the question. When I went back I didn't even view the situation as a hypothesis test. I think that is where the FRM can get you, you will think of a way to answer the question but there is the desired "correct" way to answer the question.
 

davidhn

New Member
Some more:

21/NIRP / not passing them on to retail customers / correct I think

22/MEVT / I said it can easily model any extreme event / incorrect most likely, I don't know why I went with that now..

23/Stressed RAROC / you just had to use the formula for all cases - can't remember which one but there was only one that matched / correct

24/Characteristic of distressed HF / long exposure to low credit quality / correct

25/lowest MVaR / beta was given and VaRptf/ptf value is a positive constant so just go with beta / correct

26/Asset allocation contribution / +0.28% I think - the highest value / correct I think

27/Impact of electronification / makes it advantageous to slice large orders into smallers one / could be correct but the formulation was a bit tricky

28/HFT intraday risk qualitative question on best practice / can't remember / unsure

29/BI vs standardized / I was confident saying that cap requirement would be higher under TSA, as the lowest business line coefficient is 12% as opposed to the 11% of Business Indicator, but after re-reading it I got confused and might have mixed things up / unsure

30/CCS exposure profile / FX Fwd + IRS / correct

31/Fama-French / the portfolio had a negative HML bias so expected to underperform as value outperforms growth / correct

32/New method of calculating VaR had suspiciously equal results for 99% VaR and 97.5% ES / possibly wrongly assumed normal distribution since 99% VaR = 97.5% ES for normal distributions / unsure as the formulation was tricky

33/Autocorrelation / went with educated guess / unsure

34/NSFR / unsure

35/Regression hedging / 820million or whatever the highest answer was - do not forget to adjust by beta / correct

36/Marginal and annualized PD / went with the 1 - T-th root... formula / looked right

37/Liquidity adjusted var / it was EXOgenous since we have no material impact on prices + use mean+factor*vol / correct

38/Stress loss / difference between stressed loss and "normal" loss / correct

39/40/2 questions on unsmoothing returns / I went with "no effect on expected returns" as I mistakenly thought it would simply increase volatility / wrong

41/What happens to CCP initial margin required when volatility rises / higher initial margin is required / correct

42/Another question on ES (not the 99-99.2%-99.4% one) / can't remember much more but I am fairly confident with this concept so I am gonna be optimistic / correct

43/Credit VaR calculation / UL - EL, can't remember more details / correct (straightforward)

44/45/2 questions on External debt in emerging economy/ I did not fall into the traps of "the local government would depreciate" (it says in the curriculum that precisely they would rather default than depreciate home currency) and "SMB will receive more lending than large companies" (the opposite) / educated guesses but not 100% sure..

46/A wierd (IMO) question on losing value with futures, on financials, oil, and corn - did not really understand the underlying logic / went with oil / guess

47/Stratification vs linear vs screening / I think I said LC does not take into account T-costs / probably wrong

48/Third line of defense / independent review / correct

49/I think a surprisingly easy one on cumulative PD / used 1-exp(lambda*T) / should be OK unless there was a trap somewhere...

50/CDS bond basis / ? / ?

Getting really foggy after that...
 

Hermz29

Member
Hmm, I had decided against that one because I specifically remembered the reading mentioning that a feature of Bitcoin is the lack of anonymity, and the fact that transactions can always be traced back to the individuals who were involved. I could be wrong though.
The answer has to do with anti-money laundering AML concerns.
 
Top