It's a f***ing example

 

You're confirmed +ve.

1 in a 1000 die = 0.1 in a 100, so that's 0.1% chance of dying...

...But the test is 'only' 95% successful, so you have to apply that 95% to the 0.1% i.e. 0.95 x 0.1 = 0.095%.

So you have a 'guaranteed' 0.095% chance of dying - or 95 times in every 100,000.

So, reversing it, then that's a 99.905% chance of SURVIVING.

 

Errrrrrrr....





...no

 

not a very good one

 

Okay then, it 'must' be this lol:-

5% chance of surviving, as the test is 95% accurate that you've got the ailment (therefore it's academic that the odds of dying are 1 in 1000).

 

0%


We all die sooner or later.

 

lol ^^^

But true. As it doesn't specifically state what is your chance of surviving the ailment.

 

i suspect the bankers risk models went something like this

mortgage defaults run at 2% ergo 98% of customers must be good risks

 

How specifically?

 

It was a bit like that!!


In the run up to the crunch Goldman Sachs had five days in a row where their risk models experienced 10 standard deviation events.

Each event was the equivalent of winning the lottery 21 times in a row, five times.

And yet they still used the models and didn't think - hmmm - maybe there is something awry in the Kingdom of Denmark!

 

Well for the reason I have outlined – the initial assumptions are skewed – garbage in, garbage out

However, I suspect the early risk models for sub-prime mortgage lending were both very good and very accurate in terms of risk calculation and early sub-prime providers had a valid and profitable business model

The problem is, and this to me is a fundamental flaw of capitalism (as practiced in the late 20th century) is that it just follows the line of least resistance, later iterations of the sub-prime risk model were just easy vehicles to allow “bankers” to keep the magic “money” roundabout turning, why bother checking to see if the risk models were valid, what possible advantage would that have given the banks/bankers – it is the power/allure of Market Share (apart from in the final denouement, where GS cleaned up)

Let’s face it what do bankers (generalisation alert) really know about risk, take James Dyson – he had to re-mortgage and charge his house (everything) to the banks quite a few times in the pursuit of his dream (and building a wealth creating business), how many bankers have to do the same i.e really risk their personal assets, very very few I would imagine.

 

I don't think initial assumptions are skewed. The test is designed to demonstrate that humans typically are not very good at risk assessment.

The observations I made are regarding the Gaussian Cupola correlation models and the VaR models that were used to evaluate the risk of CDOs. These were the models that were used for the creation of securitised subprime, mortgage backed securities.

The same models were used from day one - they were fundamental in enabling mortgage backed securities. People made a lot of money in the early days as the business model was founded on never ending and rapid property price rises (which of course will occur if you pump that much capital into the asset!).

I think your comment above refers to the credit scoring (FICO) models that were used in the US on the retail side of the business. These models are pretty good, but fell by the wayside for the very reason you state.

Mortgage brokers and the retail banks they worked for were betting someone else's money - so they didn't care. They became increasingly lax, and increasingly fraudulent in their origination.

 

Risk and assessment: two words I care not to combine. Far too bloody boring to waste time with.

 

Yeah 2% of mortage owners default every year, ergo 98% are good risks over the life of the mortage?

 

Agree, but they would know and awful lot about risk now if they banks had not received state aid.

Bankers should be THE experts in risk or else why does society see fit to reward them so? If they are not then they are nothing but old fashioned thieves.

 

The more you look at it the more you realise the (retail) banks didn't care.

Their agents were selling the mortgages and were being trusted to carry out the FICO credit rating and risk assessment.

The banks then passed this mortgage into the capital markets where the investment banks were providing them with AAA rated, cast iron, don't you worry your pretty little head mortgage securities.

The banks at large were merely passing the stuff through and taking a fee.

And probably half the securitisation of the market was carried out by a team of 20 people in some esoteric investment arm of a US based in London. One man judged the financial risks that were capable of bringing down the Western world's financial system.

 

The banks have a history of coming up with risk models that make the maths easy - they haven't traditionally worried about whether they are an accurate model of the real world. For example, the LTCM debacle (which involved at least one Nobel prize winning economist) was based on the assumption that stock volatility is normally distributed. If that assumption were true the kind of stock market crashes we see every ten years or so would only happen every 200 years.

LTCM came up with an algorithm for selling derivatives and then continuously hedging those derivatives in a way that guaranteed them a profit provided the price of the underlying security stayed within a certain bound either side of the strike price. Unfortunately the algorithm meant that if the price moved outside of those bounds the loses grew exponentially. Because the models assumed these movements would be very rare they massively underpriced the risk and when the stock market crashed in the late '90s they required a massive bailout.

Much of modern financial engineering follows a similar model, generating a high probability of a relatively small return (in percentage terms) in exchange for allowing a small probability of a much more significant loss. The banks then leverage up the trades to ensure the small percentage return is significant in actual cash terms. Because their models tell them the risk of significant loss is very small they kid themselves that they are not gambling. Unfortunately, as the last few years have shown, they are crap at modelling the risk - usually because they use inappropriate probability distributions or forget about whole risk scenarios altogether. CDOs is a good example of this - the risk modelling focused on mortgage defaults as independent events and almost ignored the possibility that a single underlying cause (such as widespread mis-selling or a housing market crash) could cause a large number of defaults.

If anyone is interested in this subject then I would recommend reading The Black Swan by Nassim Taleb.

 

i think what we are saying is numbers without a context are useless in determining anything

more people get attacked by sharks in 3ft of water, than at any other depth

does that tell you you are in most danger of a shark attack in 3ft of water or most people are usually in 3ft of water when attacked by a shark

 

The correlation models that were used to price CDOs used exactly that logic.

They identified a correlation between 3ft of water and shark attacks.

When the boat capsized in the deep Indian Ocean and the Oceanic White Tips came en masse they had no clue what to do. As Scud says the models work as long as nothing untoward or unplanned happens.

The 'illusion of validity'.

For books I would also recommend Liar's Poker, Big Short, Fool's Gold and Whoops.

 

so ultimately Sharks ARE to blame – I knew it

 

Durr-dum....

Durr-dum....

Dum-Dum-Dum-Dum, Durr-de, Dum-Dum-Dum-Dum...

 

Be nice to nerds. Chances are you'll end up working for one

- unknown author