Machine Learning in Trading and the Efficient Market Hypothesis
Summary
The document discusses whether using machine learning or other quantitative models in trading requires assuming the efficient market hypothesis (EMH). The responses distinguish the view that markets are broadly efficient from the possibility that temporary or not-yet-widely-known inefficiencies can persist. In that framing, a trading model searches for patterns or mispricings that may offer an edge rather than presuming that every price is always perfectly informative.
The evidence is a set of qualitative opinions, not empirical tests or a formal account of EMH. The answers also suggest that an apparent edge can be difficult to discover and may disappear when it becomes widely known. They do not establish that machine learning reliably finds profitable patterns, and they offer no guidance on validation, trading costs, or risk. The central takeaway is that a modeling technique alone does not settle the market-efficiency assumptions behind a strategy.
Key ideas
- Using machine learning in trading does not by itself imply acceptance of the efficient market hypothesis.
- Some practitioners regard markets as generally efficient while allowing for temporary or undiscovered inefficiencies.
- Quantitative models can be used to search for patterns or mispricings that may provide a trading edge.
- The responses are opinion-based and provide no empirical evidence that machine learning finds persistent profits.
- An edge may weaken once it becomes widely recognized.
Tags
Full text
# Does AI-based trading assume efficient market hypothesis? # Does AI-based trading assume efficient market hypothesis? When we use AI (machine learning/deep learning) in trading does that assume efficient market hypothesis? I know quantitative finance assumes price moves are random (efficient market hypothesis). Does quantitative analysis for trading also assume the same? Even if we assume price moves in random we can predict price of stock options.(Black-Scholes equation). ## Answer by nbbo2 (score 10) https://quant.stackexchange.com/a/79018 Most traders that I know have a complex relationship with the Efficient Markets Hypothesis, difficult to summarize. You could say that they accept a "soft version of EMH" but not the original Chicago version. They believe financial markets are indeed very efficient and therefore EMH is useful and good to know. But they believe that there may be exceptions to EMH that perhaps have not been discovered or become widely known. They tend to be very confident (you could even say a little arrogant) and believe they will be the ones to discover these inefficencies if they do sufficient research and testing. They laugh at the people who have not heard of EMH and think that just by applying a simple machine learning tool to some data from Yahoo Finance they will be able to make a fortune with an afternoon of work. They have tried that kind of thing and know it is not that easy. But they are still interested in looking for money making opportunities despite the difficulty (it is an attitude, perhaps a personality trait). If they do find something, they are very secretive about it because they know (the EMH again) that once it is widely known it will stop working. ## Answer by Sane (score 3) https://quant.stackexchange.com/a/79015 Using machine learning models (or other quantitative models) in trading does not necessarily assume the efficient market hypothesis. While some quantitative finance strategies may be based on the assumption that prices move randomly and follow efficient market dynamics, AI algorithms mostly are designed to identify and exploit patterns, trends, and anomalies in market data that go against the efficient market hypothesis (identifying inefficiencies in the market). ## Answer by Mahavir Bhattacharya (score 0) https://quant.stackexchange.com/a/79083 A common consensus among many academicians and even professional traders is that the markets are quite efficient over the long run. But there are mispricings, arbitrage opportunities, and other inefficiencies that present themselves for short time periods. The premise of using ML is essentially to recognize these edges, and capitalize on them, while trying to be on the side of the winning trades. So simply put, ML, just like any other approach to trading, or like trading itself, work on the assumption of the markets not being efficient (strongly or weakly, temporarily or permanently, would depend on the belief of the people harnessing these tools).
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.