Deep learning Examples of gaps Deep learning (DL) is analogous to induction, although based on fitting historical data. If we use DL to predict stock trends, there is a high probability of losing money, as our regulated market is constantly changing. History can repeat itself, but there must be changes. For example, US stocks are currently facing a slump. Mainly because of Federal Reserve plans to raise interest rates and tensions in Ukraine and Russia. Hence, causing flows in the money market. In this situation, DL may not be an effective tool because checking objects are still a problem.
First, DL relies on historical information.
It’s hard to infer the meaning of a word you haven’t seen, whether it’s a person or a model. If you haven’t heard of it, you probably don’t know.
Second, these names sometimes have a temporal nature, like COVID. You certainly never heard of COVID 10 years ago, and therefore it’s hard to get them in the data and learn the pattern. Now that we have these kind of names, are we going to recycle the model? It will be unreasonable because of efficiency. We can’t stop learning and new words will keep coming out.
Third, the names have a strict textual meaning. “LA LA Land” is a movie, but “XA XA Land” is not. DL can easily generalize and summarize vocabularies, but sometimes it’s not appropriate for specific nouns. Unfortunately, once the deep learning system is trained, we cannot be sure how it makes decisions. Deep learning systems may be good at recognizing pixel distribution patterns, but they cannot understand the meaning of the patterns and the reasons behind them.
Moreover, this system is easy to deceive. A single small change can generate significant deviations in forecasts. Therefore, DL can sometimes be irrational in stock trading. I think using only DL in stock trading to make decisions is an adventure: it can be advantageous, but sometimes it will lead to a biased result, which is all the more worrying as trading must be careful. and subtle.
Besides those discussed, two other major shortcomings of Deep Learning (DL) include:
1. A large number of examples are needed to solve problems. 2. Computationally intensive to train and deploy. For example, DL has a serious low efficiency problem. “Let a child get to know a dog”, “we don’t need to say ‘dog’ 10000 times, but it takes so many times for the deep learning system to learn ‘dog’. Therefore, for simple problems, it is not necessary to use deep learning.
From my point of view, the macro analysis of the market, including the development of large institutions that impact the global process, the local wars that control the distribution of resources, the policies designated by various countries, and the rise of technologies emerging, will affect the market to some extent. These are the basic reasons for determining the market trend. Then, after grasping the overall situation, we can filter the specific stock and targets in accordance with the general trend using DL.
Specifically, it is difficult to predict a “black swan event” such as COVID-19, but it is not difficult to predict monetary policy (quantitative easing) for the United States. In this situation, it is better to choose the stocks which are likely to have a higher performance in COVID, such as Amazon and Zoom using DL to find a stock trading signal.