AI in fundamental investing
Machine learning has been rapidly growing in the public markets field and I figure that in the future it will help analyze big data and be used as a better, more advanced screener. Essentially, AI will be an assistant researcher for the portfolio manager of 2030 and the investing professionals of the future will need to have data science skills in addition to the fundamentals. But I can be wrong, so I wanted to know what everyone else thinks about the relation that AI and fundamental investing will have in the future.
bump
I interviewed with couple fundamental funds who do this. Truth is at the moment, most funds have no idea how to use AI properly.
One fund actually knew what they were doing though. But how they use it is to just help make better estimations when building financial models (instead of some B-school grad fudging up some numbers). The job was really a data analyst job, more of a technician who just knows how to use models than a researcher. Not very interesting.
The future won't be much different. The way fundamental analysts think just doesn't align with highly quantitative nature of AI systems. Most likely there will be more data analysts and data scientists working side by side with fundamental analysts to help "translate" AI outputs.
Most of the AI technology used by fundamental funds will be products and services sold by firms that specialize in AI technology. That means buying things like AI document reading software, alternative data based models, etc...
The market is already going in this direction with start-ups like Kensho making NLP tools and exchange companies like NYSE, Nasdaq, IEX, CME trying to add value to the massive amount of market data they already sell.
Fundamental shops trying to be more systematic or quanty is a losing game.
Of course the catch is, there one day might be a visionary who completely shakes things up. But I doubt someone like that would even be interested in finance.
Maybe right now fundamental analysts trying to become more quanty is a failing game, but that can just be because there are not enough people out there right now who are proficient in both data science and traditional investing. This can also be happening right now because technology is not advanced enough for Ai to be able to properly support fundamental analysts (or a combination of both of these issues). Don't you think that with the rapid ascension of machine learning and data, eventually fundamental guys will find a way to merge their traditional ways with quant stuff?
Would there ever be a lot of such people? That's such a niche thing that I doubt the labor market will play out that way. If there ever are people like that, it'll probably be a coincidence.
What might happen is that fundamental analysts be required to learn some super basic data science concepts on the job.
As for me, as a data scientist working in finance, I had to pick up some accounting and finance knowledge but I was never required to learn them.
Labor markets generally tend to become more specialized over time and I don't think there are any real incentives for it to be different this time.
What do you mean by this? NLP technologies are already advanced enough for many financial companies to use AI document readers to support their work. Mayne there are some use cases of AI that haven't been tapped yet, but most of that growth is going to come from firms that specialize in AI technology. Fundamental shops will just be buyers of those technologies, not creators.
I mean yeah. They already are.
What I mean is that it's a losing game to be trying to develop this kind of technology on their own. They might hire some data analysts to make sense of some data they have in a way that doesn't require much expertise.
Otherwise they'll just purchase whatever tools are out in the market. In fact, this market is already getting pretty mature.
Where did you interview? Who do you think is doing this in the least-bad way?
It's already being used, just not in the way non-practitioners think it is.
Some places where AI is more of a buzzword than anything else include:
The distinction I want to draw is that data science does not imply AI. You'd be surprised that, even at quant firms, most of the data science being done is different forms of linear regression. At fundamental firms, it's usually just clever data cleaning, group-by, and sum. There are several reasons for this, but ultimately, the data scientists need to convince the analyst, PM, and CIO of their conclusions. Machine learning won't convince them, because they won't understand it.
With that, I do agree that financial analysts and data scientists will converge toward each other. Given how prevalent data has become in fundamental investing, the most successful data scientists can speak the language of finance, understand what drives a business, and deliver a useful result while abstracting the messy details away. And the most successful analysts won't mind getting their hands dirty with a BI tool or even SQL. They'll start munging through the data to find interesting ideas that the data scientists can spend more time fleshing out.
So basically, analysts of the future will have to know how to analyze large data and understand it but will not be required to write any code since that is going to be the job of the data scientist who will have to present the data to to pm.
there are a vast amount of data science applications for ML that are add ins for Excel, analysts and shops should consider getting acquainted with them to improve workflow. No code necessary!
I think this post is roughly right
Generally agree with the spirit of your post. Not everything is an ML problem. Most people cannot correctly identify, or are unable to formulate problems/sub-problems to in a way that allow ML to be useful.
On many occasions, the results produced can be matched by a linear model with well engineered features. I am guilty of this myself in earlier days.
100% this
I’m actually in the process of writing a research paper about this topic so maybe I can chime in. Data science is already used as an “advanced screener” for fundamental analysis, so I don’t expect anything new or groundbreaking in that regard. The real draw of ML applied to fundamentals (at least for me) would be improved predictive accuracy of credit defaults, rating changes, bankruptcies, etc.
As another poster stated, most firms don’t understand how to apply ML to fundamental analysis or have unrealistic expectations about what can be predicted and the accuracy of the models. As a side note, I think object oriented programming will become standardized sometime in the future and will be as ubiquitous as excel.
When will your research paper be done, once it is can you send it to me? I want to learn more about how data science is used in fundamental investing and figure that reading you research paper can only help.
Best case scenario, a year? I have published research (in reputable journals, not pay and prints) and I don’t usually share manuscripts or unpublished works in progress. I will be happy to post here if and when it gets published
Please do! I work am one of those quants who deal with ML and NLP professionally, and would love to see it from the academic perspective as well.
It’s encouraging that some individuals would actually find this interesting haha. Unfortunately the gathering of the data has proven to be the most time consuming and difficult aspect, models are pretty much already decided and tuning them shouldn’t be too time consuming. Imagine trying collect 30 - 40 years of bankrupt companies financials, fundamentals, and rating changes with basically a nonexistent budget.
It's already happening. Vast majority of HFs are using a process that incorporates information gleaned from the use of AI whether or not the firm itself uses AI. Great example is in equity research -- as a HF PM, when I call an analyst at a bank that's covering a security I'm interested in, they will typically have some reports that use AI/ML to evaluate certain types of risks. Insights are useful and it doesn't matter where it comes from.
Some things I've seen that were successful:
1. Vision: counting number of items from a satellite
2. NLP: reading transcripts and highlighting areas where the CEO is defensive and might be hiding bad information
3. Supervised Learning: using RF to classify trades as being part of the open auction
I've also seen it being used in amusing ways by amateurs. For example:
1. Using a NN to predict the destination of oil tankers, but the NN can't beat a model using the ship's AIS signal + a sample transition matrix.
2. Using a NN with linear activation and no hidden layers to forecast gdp (this is functionally identical to linear regression).
3. (not finance related but a well known Nature publication) Using a NN to predict earthquakes, but upon second examination the NN performed no better than logistic regression
4. Using a NN to detect if the output from an infared camera is hot/cold, when you could have hired a team of 100 Filipinos to do the same thing for considerably higher accuracy and a fraction of the cost of a US-based data scientist.
Et rerum velit minus quia sunt nobis sunt eum. Rerum et et nisi consequuntur possimus rerum at. Est ipsum veritatis velit ipsum voluptatem sequi odit. Quia reiciendis nihil sit hic similique harum.
Impedit tenetur nihil accusantium. Quisquam animi voluptates deleniti quo. Iste voluptatem sunt dolores et vitae voluptas. Dolorem exercitationem aut optio nemo.
Dolorem dicta molestiae dolores aliquam doloremque dolorem. Fugiat ea aut et laudantium consectetur natus perferendis. Pariatur nostrum excepturi et distinctio. Officiis molestiae nihil sint molestiae.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...