In this article, I would like to touch on the news information that AI agents utilize.
- Thomson Reuters Market Sentiment Index
- Scored and provided on a “per minute” basis
- Translating TRMI information into a trading methodology
- News information
- Each information movement before the policy rate was lowered
- Use of Alternative Data
- Which news information should AI use?
- in the end
Thomson Reuters Market Sentiment Index
MAiMATW utilizes the Thomson Reuters Market Sentiment Index as a news source. Refinitive’s predecessor is the financial risk division of Thomson Reuters, which we believe is one of the world’s most trusted sources of information. And their explanation of the Thomson Reuters Market Sentiment Index (“TRMI Information”) can be found at the following link.
THOMSON REUTERS MARKETPSYCH INDICES（external site）
Sources of TRMI information
TRMI Information is an information service that aggregates and indexes various news items. The various information required for indexation includes the following
- 40,000: Global News Sources
- 7,000: Blogs, stock message boards and social media sites
- 1998-present: period of providing feeds, archival storage period in the past
Scored and provided on a “per minute” basis
News information is “scored” by MarketCycle and that information is provided on a “per minute” basis. Here’s a brief explanation of the scoring method and how it works
- The direction of the information: “Is it positive or negative information?
- Overheating of information, “How many sources did you get your information from?
- The more information you get from more sources, the higher your score will be.
Unfortunately, we are unable to provide this information because redistribution of TRMI information is prohibited.
Click here if you are interested.
This book is a manual for the TRMI Information Service. The Japanese version is called “Market Psychology and Trading”.
Translating TRMI information into a trading methodology
As you can see from the previous explanations, TRMI information provides only the number of points related to the information. For this reason, it does not give you information such as “buy if it is @@, sell if it is @@”. At my mate, we examine this information carefully, and after establishing a policy for its use, we incorporate it into AI training and convert it into a trading method. Details are confidential, so we cannot share them with you. TRMI is an information source, so it is up to you to use it.
I think it’s good to look at specific data to get TRMI information and try to understand it, but it’s important to keep in mind that it’s only one example.
MAiMATE’s news information is an independently reviewed and improved version of “TRMI Information” and is available in the following four categories
Over-sold, overbought, and other information relevant to current market conditions
Market Analyst’s Market Predictions
Analysts’ estimates of the market price consist of whether it is more likely to be a buy or sell
Emotional Expressions on the Internet
Collecting Emotional Expressions on the Internet, Including A Snorting
External factors such as regulations
Aggregated information on external factors that affect foreign exchange, such as interest rates and regulations, although the news frequency is low.
Each information movement before the policy rate was lowered
Now, let’s check the price trend of USDJPY before and after the FRB’s announcement of a 0.25% cut in the policy rate on September 19, 2019 at 3:00 am and various mymate news information about “US, USD”.
After the FRB’s policy rate announcement, the price of USDJPY rose once and then fell.
Market conditions are prescient.
Prior to the announcement, we can see two strong moves in the negative direction. If there is nothing in particular, it is usually a minute move, but it seems to have been a characteristic move in the run-up to the FRB’s policy rate announcement at 3:00, so we can expect it to be somewhat useful as information to capture market events. It is up to the AI agent to decide what to do/not do with this information.
Analysts’ market forecasts that moved back and forth.
Although it is difficult to identify the characteristics of the pair, it can be seen that the market has moved downward to the left of the orange vertical line (in the information available at the time the signal was sent), but it has since moved upward. Therefore, market analysts’ forecasts for the Fed’s policy rate at 3 o’clock seemed to be wavering in one direction.
Emotional expression on a constantly moving internet
It’s very difficult to find some kind of interpretation because it looks almost like noise to the eye, but it seems to be useful information for AI agents, and the performance of AI agents using “emotional representations on the net” is not bad. Rather than capturing market events specifically, it seems to me that emotions about market events are taken in naturally, since “emotional representations on the net” operate well on a regular basis. Very interesting.
There are advance signs of external factors such as regulations
Although the news frequency is low, we can see two strong reactions before and after the Fed policy rate announcement at 3:00 a.m. For AI agents, the first strong reaction is to the left of the orange vertical line (in the information available at the time the signal was delivered), and we can trade based on whether or not to take advantage of it. It is believed that the decision may have been divided.
Use of Alternative Data
The point is that ‘My Mate provides AI agents with news information, which is different from technical information’. This information is called ‘alternative data’ and is increasingly being used in trading; AI agents are smart enough to use the information they are given if it is useful and ignore it if it is not.
Which news information should AI use?
As we’ve discussed, each piece of information has its own characteristics. So which news information should be used to create an AI?
I honestly don’t know the answer yet. What news information will work through future analysis? Future analysis of My Mate AI is needed to determine whether the combination of highly effective alternative data makes the AI’s decisions smarter and so on. We compared the results for each of the news information set up at the time of last year and found that all of the information was good and there was little difference in performance. We will continue to study and update the results.
in the end
Because TRMI information is an external service to My Mate, we believe that we need to continue to strive to understand it and use it appropriately. Since the information provided is plain and numerical, it is up to us to make the most of it or kill it.
In these days of too much information flying around, it is more important to analyze the information in order not to be at the mercy of it. We will continue to study how to improve the news information.