300 points by mlfinance 6 months ago flag hide 14 comments
johnsmith 6 months ago next
Fascinating article about using ML to predict stock market! I've been trying to implement something similar and have found XGBoost to be the most accurate.
themachinelearner 6 months ago next
Interesting! I used LightGBM and achieved good results as well. Have you tried comparing your models using cross-validation?
thequant 6 months ago next
You both make good points. Cross-validation is the way to go. What other metrics are you using to evaluate the performance of your models? I'm using MSE, R squared, and MAE.
johnsmith 6 months ago next
I've been using those same metrics and also looking at the correlation coefficient. I'm curious what frameworks or libraries you're using for your implementations?
thequant 6 months ago next
I'm using scikit-learn for most of my implementations but I'm also familiar with TensorFlow and Keras. They're great for implementing deep learning models, but might be overkill for stock market predictions. What is your experience with these libraries?
randomwalk 6 months ago prev next
Using ML for stock predictions is a lost cause. The markets are too chaotic and unpredictable, and you can't accurately model human behavior. I've been a trader for over 30 years and I've never seen it work. You should stick to the random walk hypothesis and focus on risk management instead.
johnsmith 6 months ago next
I agree with you to some extent, the markets are complex and unpredictable, but I believe that ML can help inform our decisions and help us better understand the underlying dynamics. It's not a silver bullet, but rather a tool to help us make better decisions.
themachinelearner 6 months ago next
Well said, John. ML can help us analyze the data more efficiently, but it's critical to manage our risk and consider the limitations of our models. We should view these models as complementary to our existing knowledge and intuition, not as a replacement.
algotrader 6 months ago prev next
I'm using reinforcement learning to train my agent to make trades in a simulated market. The agent learns from its past actions and adjusts its trading strategy over time. It's been performing well so far and I'm looking forward to testing it in the real market.
thequant 6 months ago next
Interesting! I haven't tried using reinforcement learning for stock predictions. What type of agent are you using, and what type of market are you simulating? Is it a continuous or order book market?
algotrader 6 months ago next
I'm using a deep Q-network (DQN) to train my agent. I'm simulating a continuous market, specifically the foreign exchange market. It's been challenging, but it's also been a great learning experience.
technical 6 months ago prev next
I prefer using technical analysis to make my trading decisions. Indicators such as RSI, MACD, and moving averages can help us identify trends and predict turning points. I've had more success with these methods than I have with ML based methods.
johnsmith 6 months ago next
I've also had success with technical analysis, but I believe that combining it with ML can lead to even better results. We can use ML to process and analyze the raw data, and then use technical indicators to generate trading signals. What do you think?