キャラクター

What are the most important features of AI algorithms?price prediction vs. position management

What are the challenges facing price forecasting?

In this article, we will focus on the contrast between “learning to predict prices” and “learning to manage positions”.
In the blog post “How to start MAiMATE, the AI trading system that grows up”, I give the following explanation about MAiMATE’s AI agent.

While many AI-based algos in the world may challenge “accurate price prediction after a certain period of time”, MAiMATE’s AI agent learns “optimal position management”.

Throughout this article, I would like to explain what exactly “optimal position management” is as clearly as possible to you.

First, to contrast “price forecasting” with “position management,” we will assume the following price data.
As you can see, there is a large decline during the period, and the key point is how to deal with this… The key point is how to deal with this.

First of all, in the case of price forecasting, we need to consider ‘how far into the future to forecast’, which is not an easy decision to make. This is not an easy decision to make. I will explain below.

We naturally tested price prediction using deep learning technology during the consideration phase of MAiMATE. Since we did not know the appropriate setting for the future prediction period, we prepared multiple prediction periods, from short to long, and examined them. However, the above-mentioned problem of the assumed operation not actually being realized was very difficult to solve, and we decided not to adopt “price prediction” in MAiMATE. In addition, when trading based on price prediction, it is necessary to separately determine “how much unrealized profit (loss) is to be fixed”, and this setting is also not easy.

There are many ways to deal with the above issues in price forecasting, and I am only explaining the “issues faced by price forecasting” in general terms.

MAiMATE’s AI algo pursues profit acquisition through “optimal position management

Next, I will explain what MAiMATE’s approach to “learning optimal position management” is. Please note that the word “alter ego” will appear frequently in the description of MAiMATE. If you haven’t read “Characteristics of MAiMATE AI – Beginner’s Guide” yet, please refer to it.

As mentioned in the past MAiMATE newsletters, the learning process for MAiMATE AI agents roughly involves the following steps: 1.
The main body creates a number of alter egos. 2.
2. each alter will be allowed to trade freely. 3.
The main unit collects the results and knowledge of each alter ego and updates its own trading style.

In other words, in MAiMATE, there is no need to set in advance “how far into the future to predict”, which was necessary in the case of “price prediction”. Also, there is no need to set in advance “how much profit (loss) will be fixed if there is an unrealized profit (loss). Everything is naturally incorporated into the “learning of comprehensive trading judgment = learning of optimal position management.

The learning process of the AI agent is shown below.

As shown in the example above, MAiMATE reviews the trade results and updates the learning results every time series. AI agent learns to comprehensively determine whether to open a new position, continue to hold a position, or close a position, and greedily continues to pursue profit. We call this learning “optimal position management. In addition, the decision to take a new position is synonymous with the prediction that the price is likely to move in that direction, and price prediction is naturally incorporated into “optimal position management. We do not know how long prices will rise (fall), but we can assume that they will rise (fall) long enough to have sufficient profit opportunities, so we take a new buy (sell) position.

However, to be more precise, the current MAiMATE only activates the AI agent once a day, so there are times when the price changes during the time when it is not activated and the best trade timing is missed. MAiMATE incorporates real-time news updates, so if the AI agent is activated in real time, it can be expected to be more agile and better able to capture changes in the situation and manage positions. I’m sure the AI agents are thinking, “We’re not serious yet! It’s too late to wake up! We hope that MAiMATE users will start training AI agents now to prepare them for the time when they will become serious. The earlier you start training your AI agents, the more phases they will experience, which will give them a big advantage in the future.

RECOMMENDED

MAiMATE Administrator

We run the official blog of MAiMATE. It provides general information about MAiMATE, including news and announcements.

RECOMMENDED

Mai mate

Trading is fun! But sometimes it's lonely. Should I set stop loss now, or wait?
If there is a partner you could trust to relieve your anxiety... Your own AI partner to fight alongside you.
That's MAiMATE.

PAGE TOP マイメイトをひらく