Just finished reading the paper Stock Market’s Price Movement Prediction With LSTM Neural Networks. The abstract attractively reads: “The results that were obtained are promising, getting up to an average of 55.9% of accuracy when predicting if the price of a particular stock is going to go up or not in the near future.”, I took the bait. You shouldn’t.
In short, the authors use over 170 technical indicators as input for an LSTM neural networks. Nearing the end of the paper I found it. The hidden sentence which the authors would do well to promote to the abstract: “When it comes to the financial results … it didn’t necessarily had the best results when compared to the baselines.” Put differently: ‘There are better alternatives than our proposal.’ It is OK to say in my opinion. It is fair, while at the same time doesn’t mean that the paper is of poor quality. For me personally this kind of honesty, if anything, suggests the reverse.
There are of course other concerns with the paper. Particularly the range of the data (2008 to 2015), specific stocks (only 5, why those 5 are chosen is unclear), and quite specific forecasting horizon (the next 15 mins). Without a good story for those choices made, and some basic robustness checks, one is quickly pressed to infer that those particular choices are motivated by the need to achieve good backtesting results.
Finally, echoing a previous post (Market intraday momentum), you would probably not publish anything you can use to generate actual income, so there’s that.