Why Stock Prediction Accuracy Does Not Guarantee Trading Performance
Summary
The discussion challenges claims about the best stock prediction accuracy by emphasizing that standalone accuracy figures do not establish a profitable, durable trading strategy. Machine-learning evaluations often assume data points are independently drawn from a stable distribution, while real trading changes market conditions: exploiting an apparent opportunity can remove it, shifting the distribution that the model learned. This feedback makes historical performance difficult to carry into live use.
The responses also raise risks such as look-ahead bias and evaluating a model on data used during development. They distinguish prediction from strategy design: a trading system also needs decisions about capital, exposure, time horizon, and risk management. No rigorous comparison of published accuracy rates or validated performance benchmark is provided, so the document does not answer which algorithm performs best. Its central caution is that reported accuracy alone says little about out-of-sample returns or ongoing tradability.
Key ideas
- Machine-learning results often rely on stable, independent observations, an assumption that markets may violate.
- Trading on a detected opportunity can change the market and weaken the pattern being predicted.
- Look-ahead bias and reuse of evaluation data can inflate reported performance.
- A profitable strategy depends on position sizing, exposure, time horizon, and risk controls as well as prediction.
Tags
Full text
# Best known performance of stock prediction algorithms # Best known performance of stock prediction algorithms I asked this question here and was directed to answer it on this stack exchange. My question is very simple. What is the best [known] performance of a stock prediction algorithm? I've seen papers with up to 76% accuracy. Has anyone [publicly] done better? PS. Here's one paper that I'm talking about http://cs229.stanford.edu/proj2012/ShenJiangZhang-StockMarketForecastingusingMachineLearningAlgorithms.pdf ## Answer by Alex Lamb (score 2) https://quant.stackexchange.com/a/21811 It's worth noting that prediction algorithms in the Machine Learning literature, if stated formally, usually come with the assumption that the data points are sampled i.i.d. from some distribution. This distribution is badly violated when the predictions are used to take actions in the real world that affect future data. For example, one might observe an arbitrage in currency prices, then trade on that opportunity, which then removes the opportunity for arbitrage. So the model will perform well, then the distribution will change, and it will stop performing well. Note that this doesn't apply to say image classification or speech recognition. Predicting that an image is of a cat won't change whether the image is or is not a cat. This is probably a big part of why Machine Learning is so successful here. On the other hand, there are lots of domains which exhibit the i.i.d. breaking feedback property. Recommendations are one such example. After Netflix recommends a movie to customers, they become more likely to see that movie, which changes the distribution of the observed data. ## Answer by Neeraj (score 1) https://quant.stackexchange.com/a/21797 All of these papers involve some kind of bias. @Alex pointed out rightly, they assume data is known in advance. Further model is tested on the same data on which it is being run. In reality, there is no such algorithm and strategy exist that can consistently outperform the market. Market is always very close to efficient. If someone able to find algorithm to predict the stock market very closely, even then this algorithm would become redundant the moment it gets disclosed. Note for cautious : Donot put your money into stock market based on the any published work claim to predict the market accurately(even for 90%). ## Answer by Marcelo Dominguez (score 0) https://quant.stackexchange.com/a/21954 Taking into account how markets works is almost imposible to predict their evolution by using any kind of algorithm... but you can use a machine learning algorithm to get positive statistical expectancy over the game so controling your capital and exposure by using using money management and risk management algorithms is possible to design a profitable trading strategy. My opinion is that there´s no good algorithms, ther´re good strategies combining prediction, money management, risk exposure, timeframes and differente exchanges.
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.