Thread by @SAH16928046

What is the Gaussian copula and how to use it to derive the joint probability of the default of two assets?

This is an interesting question, but I would like to discuss its implications and how this kind of model added fuel to the global financial crisis fire back in 2007.

Risk Management is like a Greek Tragedy, where actors laugh to express their sorrow. Hence, here what mimics laughter is the Normal (Gaussian) PDF and its assumptions.

I believe the strongest voice that emerged in the Post - Crisis years was that of Dr Nassim Taleb who heavily criticized risk management models and techniques that assume the Normality of returns, and its volatility in financial markets.

Most of the Credit derivatives and the structured products (Toxic Assets) that were financially engineered by Wall Street and the City Quants to provide a market-beating rate of returns used such techniques to pool assets having negative correlations

and a low conjoint PD.
Hence, completely overlooked how asymmetric correlations and extreme tail events(extending beyond 3 standard deviations) could complicate market and credit risk hedging in the event of a full-blown non - -normal financial crisis!

#Gaussian #Copula well and truly assumes that a financial risk statistical model will exhibit control properties on most of the days (Similar to the Gauge R & R - Repeatability & Reproducibility Analysis done in the Industrial Reliability Engineering field/s

to measure measurements/ standards) for a chosen confidence level and a risk horizon, but on the contrary, it did not live up to its reputation in backtesting.

This model was brought into use previously by Actuaries, and later infected the mindset of Quants/ Financial Engineers too!
Its rise in structured product and mathematical trading markets was idealized by the Chinese Quant Dr Andrew Lee (from China) who was working on Wall Street

Hence, it is crucial to read this article, I will add at the end of this thread, to get the hang of the full story and how it devastated the world economy back in 2007!

One big reason for the model’s failure was the lack of data samples available in the market to backtest the credit derivative losses with accuracy.
This led to complacency when undertaking model validation by the MBS Portfolio Product Developers and Marketers.

However, a successful risk/ pricing model would require not just validation but also cross-validation using training _ testing the data split technique used in data science and machine learning.
Professor Li didn't do any of that!

Please click on the URLs below to get the full story behind this financial engineering/ risk management model and its apocalyptic characteristics.
wired.com/2009/02/wp-qua…

Also read =>

pdfs.semanticscholar.org/d6a2/26e54cff1…

@threadreaderapp unroll