Statistics: Posted by Josesv — Today, 4:41 am

]]>

An engineer, a physicist, and a mathematician were on a train heading north, and had just crossed the border into Scotland.

- The engineer looked out of the window and said "Look! Scottish sheep are black!"
- The physicist said, "No, no.
*Some*Scottish sheep are black." - The mathematician looked irritated. "There is at least one field, containing at least one sheep, of which at least one side is black."

- The statistician : "It's not significant. We only know there's one black sheep"
- The computer scientist : "Oh, no! A special case!"

Wasn't he the lead guitarist with the Shadows?

Statistics: Posted by Cuchulainn — Yesterday, 2:56 pm

]]>

All the best and thanks for this super forum and information database.

Statistics: Posted by JBA — November 22nd, 2017, 11:33 am

]]>

Vega is sensitivity of an option to vol. I suspect you are thinking in terms of high vol equates to fear. People talk about VIX being a measure of fear. It's not really, it's a measure of randomness...it says more about the people who call it the Fear Index!

Here I am using Fear to mean "wanting more compensation for taking risk." (*)

Think of Randomness = Risk = Crossing the road when there's a lot of traffic. More traffic doesn't necessarily mean more fear. If you don't experience fear then you cross the road regardless. If you are afraid then either you don't cross the road (maybe that equates to selling, or less volume???) or you say "If I'm going to cross the road here you're going to have to pay me lots," and that's the same as (*).

Statistics: Posted by Paul — November 22nd, 2017, 10:06 am

]]>

I watched the documentary Quants - The Alchemists of Wall Street and at minute 06:00 you could see Mr. Wilmott drawing a graph showing lambda as a function of time at the blackboard where he is describing the upper peaks as FEAR and the lower peaks as GREED. Please find the linkt to the youtube video attached

I´m struggling in understanding the term "lambda" in this case. Is it the synonym for vega here and thus relating to option pricing?

In his paper: The Market Price of Interest-rate Risk: Measuring and Modelling Fear and Greed in the Fixed-income Markets Mr. Wilmott shows almost the same graph however the points for fear and greed are the other way around.(Negative Lambda values are FEAR and positive Lambda values are GREED)

The question that arises: Are they expressing the same - which one is then the correct one or where do they differ?

Would be great if you can give me an explenation of how to interpret this chart properly.

Thank you very much in advance!

Statistics: Posted by JBA — November 22nd, 2017, 9:37 am

]]>

Sorry, but wouldn't the above approach give you a TTC (Through the Cycle) PD, whilst for ECL you're supposed to use the PIT (Point in Time) approach?

Depends upon how one designs the estimation sample. After all, estimation needs to be done for retail portfolios for which no CDS spread is available

Statistics: Posted by rrao4 — November 21st, 2017, 2:35 pm

]]>

]]>

The bid-ask spread gives you uncertainty in your pd estimate, and I think you should then (in IFRS9 guidelines) add conservatism to compensate for that uncertainty? Is that so?

Also, I expect CDS to give a skewed view on pd, there is probably more demand than supply for cdss?

So either the mid, or if you're forced to include conservatism because of model risk then pick the side that gives the highest pd.

IFRS 9 requires that there should not be any conservatism built into the estimates. Mid point would be a better way. Typically, banks are modeling PD, LGD and EAD for both IFRS 9/CECL

Statistics: Posted by rrao4 — November 20th, 2017, 6:05 pm

]]>

Collector wrote:can I use \(\frac{x}{\frac{y}{0}}=\frac{x}{\infty}=\)0 as part of a ``proof"? or will mathematicians then see a red zero? Have they got to a conclusion on division by 0?

Is the article refereed? I think the referee will pick it up.What about in a footnote?

I was able to prove it in another way, but indirectly this gives the same for my case. (the referee is father Time, mother?, yes time will always pick it up, then drop it and dissolve it.)

Statistics: Posted by Collector — November 19th, 2017, 9:49 pm

]]>

can I use \(\frac{x}{\frac{y}{0}}=\frac{x}{\infty}=\)0 as part of a ``proof"? or will mathematicians then see a red zero? Have they got to a conclusion on division by 0?

Is the article refereed? I think the referee will pick it up.What about in a footnote?

Statistics: Posted by Cuchulainn — November 19th, 2017, 9:30 pm

]]>

Statistics: Posted by Collector — November 19th, 2017, 7:24 pm

]]>

I agree. Quant not dead but it's turning to be more engineering than science: XVA, AD, etc. To my surprise - there's still some nice research in the volatility field, I'm impressed by the book by Bergomi.

It's hard to beat the model sigma=0.2.

Statistics: Posted by Paul — November 16th, 2017, 11:31 am

]]>

I have no idea how or why DL will help PDE/PDE ... the curse of dimensionality is an

My uninformed intuition is that DL will be an auxiliary tool for Monte Carlo methods, and it's the MC part which helps against the curse of dimensionality, with DL helping to train generative models used by the MC algos.

Statistics: Posted by ISayMoo — November 16th, 2017, 11:20 am

]]>

There is a reason for this. I tried to make this point in the DL thread I created specially for cuchulainn, so he could turn it to shit. I am not advocating machine learning/deep learning per se, but I honestly believe and have some very practical experience with this, that parametric quant finance models are of extremely limited practical use in the markets. I understand that there quantitative finance is a discipline per se and it's somewhat legit for people to be interested in it for its own sake. But by and large the existence of these models is driven by and for practical use in financial markets. It's amazing just how little the average quant joe is aware of this. You just aren't going to get very far with a model that has two vols and a correlation.

yes, I remember that thread. The title was a juxtaposition of DL and PDE and a link to an article purporting to be an answer to an as of yet unstated problem description. "training data for pde" does not sound right. But who knows?

I turned off when they started talking about PDEs in 10^5 independent variables. Maybe time to get a grip.

// I have no idea how or why DL will help PDE/PDE ... the curse of dimensionality is an

Statistics: Posted by Cuchulainn — November 16th, 2017, 11:13 am

]]>

There is a reason for this. I tried to make this point in the DL thread I created specially for cuchulainn, so he could turn it to shit. I am not advocating machine learning/deep learning per se, but I honestly believe and have some very practical experience with this, that parametric quant finance models are of extremely limited practical use in the markets. I understand that there quantitative finance is a discipline per se and it's somewhat legit for people to be interested in it for its own sake. But by and large the existence of these models is driven by and for practical use in financial markets. It's amazing just how little the average quant joe is aware of this. You just aren't going to get very far with a model that has two vols and a correlation.

I am sure you can get very far with a model which has 100,000 parameters trained to a large dataset spanning over 10 years at minimum, in a way nobody really understands. I just don't know if you will like the place you will get to.

Statistics: Posted by ISayMoo — November 15th, 2017, 5:59 pm

]]>