I think MIT is doing a great service to humanity by allowing people to see such lectures online. I will donate. I hope others do too. Together we can make education a right and not a privilege.
Especially considering a conceited self-appraisal half the lecture. The fuck I care about who the teacher is. The best teachers talk about the subject non-stop, not their credentials.
Why are some people saying the video is useless. The lecturer is cool, did his best to simplify key concepts. i don't think they expected him to give a rigorous/detailed lecture. chill - and enjoy!
Super interesting and informative teacher! I am a non-market risk professional at a global investment bank and loved every minute, thanks Ken hope you are well!
Mit VaR course conducted by a top speicalist from the industry... Just wonderful. Great uni and high class lecturer. Props for opening the course for outsiders like me. 👍 Ty mit and Ty lecturer Ken
I have to say a big THANK YOU to MIT for providing these lectures. Im currently doing a math/physics double major in NewZealand but I really want to become a quant. This has really opened my eyes to financial markets and the math involved. I am honestly at peace with all the mathematics as I understand most of it. For statistics however, I have done some physics labs that make the statistical concepts in this course reminiscent of the physics labs I have done but I dont know it to much detail; so i guess im gonna have to self study statistical theory.
Hi, I am a french student majoring in financial markets and I would have loved to have you as my teacher. You look so passionate, I envy your students. Thanks for the video!
All, there is an error on slide 40. The formula for the portfolio variance should be: =MMULT(MMULT(A2:C2,E2:G4),TRANSPOSE(A2:C2)) then ctrl+shift+enter. Despite the small error, spectacular video, thanks for posting MIT! Kenneth Abbott, you are a great teacher, thanks for sharing your knowledge!
@Ken Abbott Haha, totally understand. Would you mind sending me your email? Would really appreciate having a contact like yourself for topics like this. Best!
I know this is late, but this greatly helps me understand the various scenes in Margin Call where they were discussing the potential upcoming losses. Not just that they were about to lose a huge pile of cash, but how they figured out that it was about to happen. Had no real clue or understanding of what VAR numbers meant.
Also FORTRAN & COBOL coder using cards. USAF required that I be able to read blank Hollerith Cards at 1 card per minute, Baudot paper table at 25 characters per minute. Am another of the ancient ones.
@@kennethabbott8248 Mr. Abbott, where can I find related papers/journals from which I can read more about VAR, Portfolio, Options, Derivatives, etc related to finance? I believe that getting a good in-depth know-how of these things would be beneficial in the long run, atleast for me.
Amazing lecture! Thanks MIT Just one thing, what he said at 0:28:45, "the likelihood of a move [in a fat tailed distribution] of 2.33 SD is more than 1%". Isn't it the other way around? It's fewer observations outside of 2.33 SDs in a fat tailed distribution than it is in a normally distributed.
@@kennethabbott8248 Doesn't that depend on how the fat tailed distribution look? The probability for a extremely large loss are indeed higher in a fat tailed distribution. But it can still be more likely for a 2,33 sigma loss in a normal. For example, 2 distributions, one normal and one fat tailed with the same mean and variance. Both have 1000 observations. In the normal it will be approx 10 events outside 2,33 sigma, but in the fat tailed it can in the extreme case be 1. If that observation is far enough out in the tail, like 15-20 sigma, then the distributions can still have the same variance. Thus the probability of a 2,33 sigma loss is smaller in the fat tailed. Or am I missing something?
@@moosecapital8886 Technically, of course, you are correct. The fact that the two cumulative distribution functions cross each other demands that they cross a second time since both of the associated density functions must integrate out to 1.0. In practice, in a diversified portfolio, the VaR at 99% and 95% are always higher for HS than for VCV or MC. (At least I have never seen any instances of an empirical VaR at these levels be less that that implied by the normal.) Ive seen the crossover between 80% and 95%, typically around 90%. My $0.02.
47:38 that position vector should be read a column vector in order to conduct x'Ex..otherwise if some people read it as a row vector, you can't do x'Ex because (nx1) cannot multiply a (nxn) matrix..
Slide 23 - there is something maybe wrong there? Cause if first says that Credit Spread takes recovery into consideration. Then it's said that PV01 = CSPV01 only if recovery = 0. I think this is not the case (if we define credit spread as considering the LGD as well: in this case any 1bs change in both risk-free rate and Credit Spread will result in a same PV01 = CSPV01. Can anyone check this?
Awesome lecture! Lucky MIT kids. I have a master degree in risk management. I have learnt almost all of the stuff, but I enjoyed it a lot! I love how it refreshes my memory and links my knowledge to practical world. And I liked the jokes haha.
Hi There Lecture is really very good as it is presented by experience quant. Is there any possibility to get the Excel spreadsheet presented by Kenneth? Thanks, KP
Krunal Patel Sorry, the course materials do not include any Excel spreadsheets... but the course lectures notes do have a number of data tables on the slides that might be of some help? See the course on MIT OpenCourseWare for the course materials at: ocw.mit.edu/18-S096F13
He has a serious problem with Mother Theresa huh? Jokes aside, I find it a really enjoyable lecture and one of the few ones that don't need 1.5x playback. Thanks and best wishes!
Hey guys! I´m from Argentina and i´ve got a question about the adjustemnts in the covariances matrix for this methods. (0:32:07) ¿What are the best approch to expressing the covariance matrix in basis points with high inflation economies? I mean, if i´ve got a yield series ¿can i only apply absolute changes for the returns? Or, ¿maybe, i have to took the log returns and then transform the covariance matrix in basis points? Thank you so much!
Remember VaR always assumes stationarity. For yields, use % change (or log change) and apply that to the most recent yield you have: position * PV01 * close * s * 2.33 *100, where s is the stdev of the yield changes.
@@kennethabbott8248 thank you so much for your answer! Really. I was referring to the units. I’ve read that, as PV01 is in dollars i have to re express the covariance matrix in bps.
@@lautaropintos1 The covariance is always that of the returns (% changes or log changes). You make the units adjustment in the position vector X, where the variance formula is X'VX and V is the covariance matrix. BTW: very few people use this approach any more. Most banks (~86%) use Historical Simulation.
very well explained！ though he jumped too fast about the details of how he constructed the mcs. The core is to use the portfolio correlation matrix's sign vector and eign value to transform normal distributed data to be the same distribution as the portfolio correlation. Then we can generate as many as fake data as we want to simulate what's going on in the worst case.
Please: how to do the final transformations in order to obtain the correlated synthetic data! A practical explanation of that would be most useful! Thank you for this video.
Loved this class. The comments about the sign of an FX contract in the position vector really clarified some practical issues to me. By the way, is that spreadsheet on Monte Carlo avaliable somewhere? Couldn't find it in the related MIT course page.
@@kennethabbott8248 Hahaha after 25 years that's pretty damn good!!! I've only been able to make it to 5, and my fiancee dropped my course...(then me...)
There was a tax on assets in the city of Chicago in the spring of 1932. It caused people to bid the T-Bill through par. This is borne out in the Federal Reserve Bulletin of that time. There is an asterisk instead of a yield in a table of rates.
@@kennethabbott8248 thanks for replying so quick! i just thought it was funny seeing as how the lecture was done in 2013 and fast forward a few years and many gov bonds around the world are yielding negative. thanks for providing context too!
CORRECTION - From Cecchetti, The Case of the Negative Nominal Interest Rates: New Estimates of the Term Structure of Interest Rates during the Great Depression: "On December 31, 1932, the New York Times listed the yield on a 3.5 percent U.S. liberty bond as - 1. 74 percent. ...It is well known that during the Great Depression the prices of Treasury bills at auction occasionally exceeded par.."
Sure, but remember that the covariance structure that is the basis for your simulation isn't metaphysical truth, but rather an estimate. If you have few datapoint, your estimate is poor.
I think MIT is doing a great service to humanity by allowing people to see such lectures online. I will donate. I hope others do too. Together we can make education a right and not a privilege.
@@raulbartolome8389 Not all heroes wear capes, thanks man! :)
YES KNOWLEDGE FOR ALL
Agreed
Especially considering a conceited self-appraisal half the lecture. The fuck I care about who the teacher is. The best teachers talk about the subject non-stop, not their credentials.
So what? He seems an intelligent engaging and sympathetic teacher.
Some notable Timestamps:
0:06:19 Methodology
0:38:33 Covariance, Correlation & Matrix algebra
0:46:50 The big picture
1:00:51 Monte Carlo
44:40 cookie
This man is brilliant. We are very lucky. Thank you MIT for the content and the sourcing of brilliant lecturers
Best takeaway from the lecture: “Whats the difference between a bond and a bond trader? A bond matures.” 😂😂😂
🥲🤣🤣
45:45 that was a great insight for simplifying portfolio volatility calculations... wow
Six years later and this is still an amazing lecture. Thank you so much!
Why are some people saying the video is useless. The lecturer is cool, did his best to simplify key concepts. i don't think they expected him to give a rigorous/detailed lecture. chill - and enjoy!
He might not have a PhD, but he is an amazing lecturer. Natural born teacher.
Thank you. I'd be a lesser teacher if I had a PhD. My value-add proposition is to explain how things are actually done.
It seems to me that he can empower people and has a lot of knowledge. So the PhD is just a title not a necessarily proof of knowledge.
@@kennethabbott8248 well-said!
he actually has more technical knowledge about stats and econ than most of the phds out there.
Super interesting and informative teacher! I am a non-market risk professional at a global investment bank and loved every minute, thanks Ken hope you are well!
I am well, thank you very much. I'm glad you enjoyed it.
Excellent teaching, well done! Been through a Bachelor and now doing my Masters', never found anyone being this easy to understand.
I'm glad you liked it.
Along with explaining the concepts in simple and effective manner he gave life lessons too! Thanks a lot Sir!
wonderful lecture. I can't believe this is available to general community. Thank you so much, MIT!
I love that "can I have a piece of that cookie" hahahaha
what a human moment
The stats teacher we all wish we had. Thanks for the upload, MIT.
Mit VaR course conducted by a top speicalist from the industry... Just wonderful. Great uni and high class lecturer. Props for opening the course for outsiders like me. 👍 Ty mit and Ty lecturer Ken
Happy to be of use. Let me know if you have any questions.
I have to say a big THANK YOU to MIT for providing these lectures. Im currently doing a math/physics double major in NewZealand but I really want to become a quant. This has really opened my eyes to financial markets and the math involved. I am honestly at peace with all the mathematics as I understand most of it. For statistics however, I have done some physics labs that make the statistical concepts in this course reminiscent of the physics labs I have done but I dont know it to much detail; so i guess im gonna have to self study statistical theory.
hey oscar ,how ıs ıt goıng your beıng quant adventure?
Who is still bumping this 2019?
Why do you ask?
can't believe they are having such great courses online for free! Courses are very good.
Hi, I am a french student majoring in financial markets and I would have loved to have you as my teacher. You look so passionate, I envy your students. Thanks for the video!
Very informative lecture and his sense of humor is on another level of graetness.
2 Years later and this is still an amazing lecture.
Thanks
@@kennethabbott8248 No, thank you. Amazing lecture!!!
It’s 2023 but his lecture is still useful and plus points to Mr. Abott’s quirkiness and witty jokes hahahaha
This professor has an awesome way of teaching.
I wish this man was my mentor as I begin my journey. Great attitude and lecture.
I'm happy to provide any help I can.
@@kennethabbott8248 hi Ken ı realy need help about my life ı have to ask questions How can ı connect to you
So glad I saw this. This opened my mind wide open about finances. I need to study this more
Incredible teacher! I just started my career in Financial engineering attribute and I cant agree more that never trust the data from others!!!
Thanks.
Same career
his energy is infectious.
somebody give this guy one more cookie!!!!
XD
Thanks a lot to MIT and Mr. Abbott! A very lively and essential lecture.
I wish they had a whole lecture series with this instructor.
I do a whole lecture series every year at Baruch in the MFE program. I teach undergrads, as well. Come sit in!
@@kennethabbott8248 Thanks! And thanks for the great work you put out!
What an attitude....cool nerd!! Need more teachers like him
All, there is an error on slide 40. The formula for the portfolio variance should be:
=MMULT(MMULT(A2:C2,E2:G4),TRANSPOSE(A2:C2))
then ctrl+shift+enter.
Despite the small error, spectacular video, thanks for posting MIT!
Kenneth Abbott, you are a great teacher, thanks for sharing your knowledge!
@Ken Abbott Haha, totally understand. Would you mind sending me your email? Would really appreciate having a contact like yourself for topics like this. Best!
abbottk (at) post (dot) harvard (dot) edu.
Unbelievable ! based on this video. i passed a counterparty risk job interview.
I'm glad it was useful for you.
Your tips help me become a better trader. Thank you very much for your time and effort!
So relaxing to hear everyone coughing, miss these old days
if all my lecturers were as good as you, i would be going to every lecture happily
Wao. This is an awesome lecture on risk management. I love your sense of humor. Thank you sir.
Thanks. My wife thinks I'm just ok, though.
post more videos of Mr. Abbott
This man is amazing. I would be honored to work with him.
Thank you for making this available to everyone.
I know this is late, but this greatly helps me understand the various scenes in Margin Call where they were discussing the potential upcoming losses. Not just that they were about to lose a huge pile of cash, but how they figured out that it was about to happen. Had no real clue or understanding of what VAR numbers meant.
Also FORTRAN & COBOL coder using cards. USAF required that I be able to read blank Hollerith Cards at 1 card per minute, Baudot paper table at 25 characters per minute. Am another of the ancient ones.
Fantastic, fantastic lecture.
I'm glad you liked it.
@@kennethabbott8248 Mr. Abbott, where can I find related papers/journals from which I can read more about VAR, Portfolio, Options, Derivatives, etc related to finance? I believe that getting a good in-depth know-how of these things would be beneficial in the long run, atleast for me.
This lecture and the lecturer are amazing! Thanks for sharing it.
I love this guy, at 50.10 he had me dying laughing. Kenneth I will bring you an entire box of cookies to come sit in on one of your lectures.
Let me know when you're in NYC. You can sit in.
@@kennethabbott8248, Great lecture! definitely useful even not for quants (CS grad from Israel here)
@@kennethabbott8248 Sorry but at 52:15 that bit about Abe Lincoln and the Internet just fell completely flat 😂
@@HelloThisIsAnon Yeah. That one didn't go over.
I have spent so many years at several universities, I have never seen something like thatbefore!! I laughed out loud! That was genuinely funny!!!
(2020) thank you so so much for all those explanations ❤️
I'm glad you found it to be of use.
The instructor is very down to earth. I like it.
Very helpful for people who are trying to self study this stuff. Thank you so much for the lecture and the anecdotes.
I'm happy to be of service.
"Are you taking something for that?"
Me watching this in 2021.....
I wish attending one of your great, rich and well presented classes, great work
The lecturer is so good
Amazing lecture! Thanks MIT
Just one thing, what he said at 0:28:45, "the likelihood of a move [in a fat tailed distribution] of 2.33 SD is more than 1%". Isn't it the other way around? It's fewer observations outside of 2.33 SDs in a fat tailed distribution than it is in a normally distributed.
No. Leptokurtosis implies a higher probability of larger losses.
@@kennethabbott8248 Doesn't that depend on how the fat tailed distribution look?
The probability for a extremely large loss are indeed higher in a fat tailed distribution. But it can still be more likely for a 2,33 sigma loss in a normal.
For example, 2 distributions, one normal and one fat tailed with the same mean and variance. Both have 1000 observations. In the normal it will be approx 10 events outside 2,33 sigma, but in the fat tailed it can in the extreme case be 1. If that observation is far enough out in the tail, like 15-20 sigma, then the distributions can still have the same variance. Thus the probability of a 2,33 sigma loss is smaller in the fat tailed. Or am I missing something?
@@moosecapital8886 Technically, of course, you are correct. The fact that the two cumulative distribution functions cross each other demands that they cross a second time since both of the associated density functions must integrate out to 1.0. In practice, in a diversified portfolio, the VaR at 99% and 95% are always higher for HS than for VCV or MC. (At least I have never seen any instances of an empirical VaR at these levels be less that that implied by the normal.) Ive seen the crossover between 80% and 95%, typically around 90%. My $0.02.
@@kennethabbott8248 That makes sense, thanks a lot for the reply Kenneth. Really enjoyed the lecture
0:31:53 Crazy... 9 years later and negative yields are common all around the world
The Cornish-fischer expansion technique is also very useful and practical for VaR
Thanks Prof. Kenneth for this amazing lecture!
I guess I will be able to replicate it myself, thanks Mr Kenneth for the keys of the KINGDOM
me still watching this video in 2024...
😂
Me toooooo
Way better than LSE finance lectures
47:38 that position vector should be read a column vector in order to conduct x'Ex..otherwise if some people read it as a row vector, you can't do x'Ex because (nx1) cannot multiply a (nxn) matrix..
Yes. That is correct.
@@kennethabbott8248 Thanks Ken. I wish there would be your 8-hour course! I learned a lot from your condensed lecture!
Slide 23 - there is something maybe wrong there? Cause if first says that Credit Spread takes recovery into consideration. Then it's said that PV01 = CSPV01 only if recovery = 0. I think this is not the case (if we define credit spread as considering the LGD as well: in this case any 1bs change in both risk-free rate and Credit Spread will result in a same PV01 = CSPV01. Can anyone check this?
It's a fair point, but I think you mean PV01=CSPV01 IFF LGD=0, not recovery. I have wrestled with this issue for many years.
Thanks for this video it gave me the basic background I need to understand VaR in my Financial Engineering Module.
Goid vid. Like the experience and not theory perspective of bad data.
i never seen a stat finance professor funny enough to have his own sitcom, i tell u wat
Thanks MIT ...Harvard you should take example of thé best education non profit approch
At 30:30 you mentioned graphing the data - what do you look for in the graph? If prices are true random walks then what good is graphing the data?
should the x bar in the top eqn at 19:30 be mu? that is to say sigma^2 = E[(X_i - mu)^2]?
This is a cool lecturer , when I get money , I will donate.
Awesome lecture! Lucky MIT kids.
I have a master degree in risk management. I have learnt almost all of the stuff, but I enjoyed it a lot! I love how it refreshes my memory and links my knowledge to practical world. And I liked the jokes haha.
This lecture is amazing
The guy coughing at 29:03 reallllly pissed me off lol
Hi There
Lecture is really very good as it is presented by experience quant.
Is there any possibility to get the Excel spreadsheet presented by Kenneth?
Thanks,
KP
Krunal Patel Sorry, the course materials do not include any Excel spreadsheets... but the course lectures notes do have a number of data tables on the slides that might be of some help? See the course on MIT OpenCourseWare for the course materials at: ocw.mit.edu/18-S096F13
He has a serious problem with Mother Theresa huh? Jokes aside, I find it a really enjoyable lecture and one of the few ones that don't need 1.5x playback.
Thanks and best wishes!
Thanks. I clearly can't please everyone, but if I can make the material clearer for a few, I'm satisfied.
This guy is hilarious, and doesn't use jargon. Keeps it very close to application and explains things well.
This has over 450,000 views. Ken Abbott, there is a lot of demand for an entire course if you are up for it!
I've been teaching this material as part of a risk for many years. I now teach it at Baruch and Claremont and co-teach it at NYU.
Can I have a piece of your cookie - this guy is a legend
Hey guys! I´m from Argentina and i´ve got a question about the adjustemnts in the covariances matrix for this methods.
(0:32:07)
¿What are the best approch to expressing the covariance matrix in basis points with high inflation economies? I mean, if i´ve got a yield series ¿can i only apply absolute changes for the returns? Or, ¿maybe, i have to took the log returns and then transform the covariance matrix in basis points?
Thank you so much!
Remember VaR always assumes stationarity. For yields, use % change (or log change) and apply that to the most recent yield you have: position * PV01 * close * s * 2.33 *100, where s is the stdev of the yield changes.
@@kennethabbott8248 thank you so much for your answer! Really. I was referring to the units. I’ve read that, as PV01 is in dollars i have to re express the covariance matrix in bps.
@@lautaropintos1 The covariance is always that of the returns (% changes or log changes). You make the units adjustment in the position vector X, where the variance formula is X'VX and V is the covariance matrix. BTW: very few people use this approach any more. Most banks (~86%) use Historical Simulation.
@@kennethabbott8248 thank you so much!! I’ll think about it 😊
What an amazing teacher. Great lecture
Excellent. I am so thankful for this video.
Great Content, Thank you MIT.
A true teacher, awesome indeed!
Thanks.
very well explained！ though he jumped too fast about the details of how he constructed the mcs. The core is to use the portfolio correlation matrix's sign vector and eign value to transform normal distributed data to be the same distribution as the portfolio correlation. Then we can generate as many as fake data as we want to simulate what's going on in the worst case.
@Ken Abbott lol yeah totally understood! The proof is easy and will be leave as an exercise, said by professor
Good. I am ready to play my part in next financial crisis now.
You can use polar coordinates for correlation
Please: how to do the final transformations in order to obtain the correlated synthetic data! A practical explanation of that would be most useful! Thank you for this video.
This might be useful: th-cam.com/video/QCqsJVS8p5A/w-d-xo.html
Loved this class. The comments about the sign of an FX contract in the position vector really clarified some practical issues to me. By the way, is that spreadsheet on Monte Carlo avaliable somewhere? Couldn't find it in the related MIT course page.
53:00 Bloomberg data error
Amazing Lecture!
43:12 - should the decomposition of covariance matrix be E * Lambda * (E transpose)?
@Ken Abbott Wow! I really didn't expect the explanation from the lecturer! Thank you so much!
Thank you! I could listen to you for ages! :)
My wife says I'm only "better than average". That's all I got after 25 years I guess.
KA
@@kennethabbott8248 Hahaha after 25 years that's pretty damn good!!! I've only been able to make it to 5, and my fiancee dropped my course...(then me...)
What's the difference between a bond and a bond trader? The bond matures
Watching this in 2020 while in iso
31:40 lol he hints that there's been times when yields might go negative....
There was a tax on assets in the city of Chicago in the spring of 1932. It caused people to bid the T-Bill through par. This is borne out in the Federal Reserve Bulletin of that time. There is an asterisk instead of a yield in a table of rates.
@@kennethabbott8248 thanks for replying so quick! i just thought it was funny seeing as how the lecture was done in 2013 and fast forward a few years and many gov bonds around the world are yielding negative. thanks for providing context too!
CORRECTION - From Cecchetti, The Case of the Negative Nominal Interest Rates: New Estimates of the Term Structure of Interest Rates during the Great Depression: "On December 31, 1932, the New York Times listed the yield on a 3.5 percent U.S. liberty bond as - 1. 74 percent. ...It is well known that during the Great Depression the prices of Treasury bills at auction occasionally exceeded par.."
Does somebody have the link to the spreadsheet with his Monte Carlo simulations?
everytime there is a lecture,the same old professor will be seating in the seats,he must like studying a lot
At 1:11:14. I think [INAUDIBLE] can be WATFIV.
Amazing teacher
1:18:00 if you only have 100 points of data and you run a Monte Carlo wouldn't it be too similar and not include long tails?
Would you be able to run a Monte Carlo sim for expected returns on a small sample set of performance data? Thanks
Sure, but remember that the covariance structure that is the basis for your simulation isn't metaphysical truth, but rather an estimate. If you have few datapoint, your estimate is poor.
what a great professor!!
Thanks
Great lecture, Mr Abbott! Have a cookie!
1:11:38 - why did it get kinda awkward there?
I love this guys, incredible amazing