I have a PhD in ML but recently found economics super interesting. Trying to figure what the best way to become an expert in Causal Inference and economics. Are there good courses and research groups that I can follow that Blind folks can recommend?
Thank you. Could you please elaborate on why you say that ? There have been really cool advancements lately. Meta learners, double machine learning. Causal AI will be a disruptive.
There is a temporal dynamics associated with responses. It’s not like a clear cut I do this, and the response exploded by this much after this intervention. Most times, there are confounding variables that you probably haven’t accounted for and they could also change over time. So, trying to associate a cause to an event over a longer horizon is always going to be viewed this skepticism and rightly so. On the other hand, if you look at other responses from Google where you are maximizing short term impact, you might see the lift / fall and clearly associate variables for that horizon.
If you do a solid experiment or even quasi experiment your critique matters less. You're definitely right that there's temporaral effects that pop up if you don't have perfect information though. That affects any type of epistemologic approach though.
As a relative novice in this field, I had found casual inference for the brave and true to be a great read.. then had skimmed the r-learner, double ml etc papers… also open source lib: https://github.com/uber/causalml… working on different things now but it was really cool to read as an eng
Causal Inference for brave and true is hilarious and insightful at the same time. Does Uber actually use CausalML package internally or have they upgraded to something better?
I love the frequent mixup of "casual" and "causal" in this thread.
Fixed. 🙏🏽
Read Imbens' review of Pearl's book if you haven't already, will give you a good comparison point of different approaches to causality. As to who uses this stuff, you're looking for analytics research type teams. Idk what the job title would be. Uber and MS for sure had teams that did this stuff. But take this with a grain of salt, I'm about two years out of date on both adoption and research at this point.
I'm involved with causal inference work at MS. Happy to discuss more.
Dowhy by microsoft. Try contributing to it. Join their slack they welcome controbutors
That’s a great idea. They are building a great stack with dowhy and econml.
I have a PhD in Econ with research in CausalML. Amazon has a whole job family dedicated to this - Econ RFCA, where I used to work for 5 years. Almost all companies will have a smattering of these positions. Honestly, at this point in my career I see ways to move - the "enhance AB testing" route where you enhance AB testing with CausalML, or the Product management route - then you basically become a PM with the most stat, econ, and ML experience. I see a lot of value for the latter.
Econ is dead. There is more causal inference research in pure Stats than Econ
I think you will be amazed if you actually know what’s going on in Econ research
Use fucking all the data to predict your countries economy and start a blog or site to invest to make some profit, gain damn patrons
Do phd in econ
Not sure that can happen. Already have a PhD in ML.
Dead field
Can you expand on this? Is this because there are better methods for estimating things causal inference usually would (A/B tests, ML explainability), or just that there is no new innovation happening?
As a financial economist, you are trained to look for explanations. However, even the best financial experts cannot always explain why the stock price moved yesterday. This is because the stock market is a complex system with many interconnected factors. Trying to find a single cause for a stock price movement is often futile. So ML is better at predicting events, making it useful for most cases.