Probabilistic Programming in Quantitative Finance by Thomas Wiecki, PhD
Вставка
- Опубліковано 1 жов 2024
- There exist a large number of metrics to evaluate the performance-risk trade-off of a portfolio. Although those metrics have proven to be useful tools in practice, most of them require a large amount of data and implicitly assume returns to be normally distributed. Bayesian modeling is a statistical framework that allows great flexibility in modeling financial returns as well as risk metrics. In addition, uncertainty of these metrics can be directly quantified in terms of the posterior distribution.
Thomas will briefly provide an overview of Bayesian statistics and how Probabilistic Programming frameworks like PyMC can be used to build and estimate complex statistical models. He will then show how several common financial risk metrics like the Sharpe ratio can be expressed as a probabilistic program. Using real-world data from anonymized algorithms running on Quantopian (www.quantopian...), he will demonstrate how the normality assumption can strongly bias the Sharpe ratio and how heavy-tailed distributions can remedy this problem.
Disclaimer
Quantopian provides this presentation to help people write trading algorithms - it is not intended to provide investment advice.
More specifically, the material is provided for informational purposes only and does not constitute an offer to sell, a solicitation to buy, or a recommendation or endorsement for any security or strategy, nor does it constitute an offer to provide investment advisory or other services by Quantopian.
In addition, the content neither constitutes investment advice nor offers any opinion with respect to the suitability of any security or any specific investment. Quantopian makes no guarantees as to accuracy or completeness of the views expressed in the website. The views are subject to change, and may have become unreliable for various reasons, including changes in market conditions or economic circumstances.
"quite a few statisticians I know will be satisfied with that and say OK we are done here". You want to sell yourself and your technology, fine, but saying that "for the rest of us living in the real world" ? Are you implying that Hastie and Tibshirani do not know what are the right methods for the right situation ? That the whole discipline of Statistics is bogus academic circle with little use in the real world ?
+Thomas Wiecki Thanks for clarifying. I think my reaction is mostly because I have fought off many mis-applications of machine learning / data science in my career. It is usually not that hard when you have a strong quant culture and clear veto rights. However, lately I realise that I have to explain even the most basic things like "simulation of X+ Simulation of Y is not equal to simulation of (X+Y) because X and Y are not independent". People who make these mistakes are those who do courses after courses in Data Science online which show them a few recipes of what to do, but do not tell them what *not* to do. In current job market, since those skills are harder to test, there is no reason for the new generation to learn the fundamentals anymore (for example in a statistics course). It is a sad state but so it is. When someone like you, who a lot of people follow talks about Statistics, I think there is a danger of people taking you quite literally. Thanks for clarifying.
+Thomas Wiecki Here is an article I wrote on it long time ago. www.linkedin.com/pulse/why-re-branding-your-skills-good-society-shobhit-verma
@Thomas Wiecki You're great figure for us, by just reading this conversation I realized what a humble human being you are! Thanks sir!
Very good insights that I've found in my work as well. Is there a recommended book for Probabilistic Programming that you would recommend?
Oh and there are methods to estimate uncertainty of estimates in statistics. I love Bayesian statistics but I do not find this pattern, of labeling statisticians as impractical problem solvers, amusing.
very good presentation. really found it interesting and illustrative, especially the final example with the sharpe ratios. is there a way to get a copy of the slide, or jupyter notebook, so that we can walk through the examples ourselves? thanks
Hi James,
We don't have a copy of the notebook available on our platform as it’s an older tutorial. We do however have plenty of other notebooks and sample algorithms on our platform that you can walk through to get a more hands-on experience. A good one to get started with is the Getting Started Tutorial, which can help get you up to speed on the platform before you dive into more complicated concepts. You can find the Getting Started Tutorial here: www.quantopian.com/tutorials/getting-started. We also have a variety of different lectures and other educational materials that you can find at www.quantopian.com/contest/resources and here on our UA-cam channel. In addition to all the free educational material on our site, we also have community forums, where you can interact with other members of the community, ask questions, and find some new ideas. You can access Quantopian's forums at www.quantopian.com/posts. Hope that helps!
Did you also place a prior on the extra parameter when using T distribution?
I have the same question, but it will probably be obvious if we just give pymc3 a try
Great talk
It worries me that the model is very sensitive to hyper-parameters / human error. Like you change hyper-parameter of the model (different distribution) - and got completely opposite results.
Yeah but since the second model really does seem to fit the data well, any further changes will probably lead to much smaller changes in the results (at least that's my intuition about this)