Friday, February 23, 2024
HomeTechnologyAchieving one-year market forecasts from news text information!

Achieving one-year market forecasts from news text information!

xenodata market lab., which provides the cloud service “xenoBrain” that specializes in economic forecasting, uses multiple models such as deep learning from text information such as news to predict market conditions such as the Nikkei average and crude oil prices. Industry statistics, including current production statistics and major statistics for each industry, and about 25,000 indicators have been forecasted up to one year ahead, and are now available as a new function within xenoBrain: market forecasts.

“xenoBrain” is an AI cloud service that predicts various economic information by analyzing economic news, corporate disclosure materials, and statistics with its own AI. 

By analyzing the connections between economic events and predicting economic topics and statistics such as corporate performance, material prices, and industry demand, we support the management decision-making of business companies and the operational efficiency of financial institutions. 

We collect and analyze highly reliable information through partnerships with major media outlets such as Dow Jones and Jiji Press, and with Teikoku Databank.

market

The market forecast released this time is the first in Japan to predict quantitative data (market demand, material prices, etc.) by learning 3,000 news items and more than 30,000 statistical data per day with multiple models such as deep learning. It is an economic forecast by the general-purpose economic forecasting model of .

Table of Contents 

  • Main forecast target of “Market forecast” function
  • Main use cases of the “market forecast” function
  • Features of the “Market Forecast” function
  • About prediction accuracy
  • How the Market Forecasts feature works
    • Step1 Analysis of news data
    • Step2Extracting features from news
    • Step3 Prediction based on the extracted features

Main forecast target of “Market forecast” function

<Domestic/Overseas Macro> Prices, Bank of Japan Tankan, Trade
Statistics, Balance of Payments, Labor Statistics, Business Trends Survey, Public Finance, Family Income and Expenditure Survey, Land Prices , etc. Chemicals, pharmaceuticals, petroleum products, ceramics, steel, non-ferrous metals, electrical machinery, machinery, transportation machinery, etc.

Prices, sales volume, production volume, etc. <Market conditions> Crude oil prices, wheat, precious metals, cotton, food, grains, etc. <Finance> Interest rates , bond prices, stock indices, futures prices, exchange rates, options, swaps, corporate bonds, bank transactions, etc.

Main use cases of the “market forecast” function

Corporate Planning Division By easily grasping raw material prices and product demand trends, it can be used for raw material procurement, decision making and budgeting for capital investment, and preparation of market outlook materials for use in management meetings.
Procurement Planning Department It can be used as a material for examining prices and procurement timing when planning strategies for raw materials and parts procurement.
Sales department When formulating a business budget, by confirming the outlook of each industry for about one year with objective figures, it is possible to formulate a highly accurate budget plan or use it as a basis for discussions.
Market department You can use it as a material for investment decisions by checking the trends of the stock market and commodity market and grasping how the market will move in the next year or so.

 

Features of the “Market Forecast” function

General-purpose prediction that can predict indicators of various industries with a single mechanism by extracting feature values ​​from news data, which is unstructured data and difficult to use for quantitative data prediction. realizing the system.

<Advantages of extracting feature values ​​from news>
Advantage 1: Versatile predictions are possible. It is possible to make predictions by inputting economic phenomena related to all industries such as the automobile industry and the semiconductor industry. 

General-purpose forecasting is possible because there is no need to prepare training data or adjust model parameters for each forecasted industry or forecasted indicator.
Advantage 2: Materials can be considered earlier than
numerical data.

One of the characteristics of news itself is that it often appears in the market as materials earlier than it appears in numerical data (for example, news that an automobile manufacturer has announced a production cut).

is on the market faster than the number of cars sold or the demand for auto parts is available in the statistics.) Therefore, it is possible to make more accurate predictions than when learning and predicting only from quantitative data such as statistical data.

About prediction accuracy

xenoBrain performs accuracy verification on all statistics, and provides statistics (25,000 statistics) that have been determined to have a certain level of accuracy or higher. The prediction accuracy verification results for the 6,000 most important indicators after 6 months are 11.315 mean square error, 0.768 correlation coefficient, and 0.120 mean error rate.

How the Market Forecasts feature works

Step1 Analysis of news data

Over 3,000 economic news articles delivered daily from approximately 160 domestic economic media outlets are summarized and structured data (*1) are generated using our unique natural language processing.

Step2Extracting features from news

From the structured data analyzed in Step 1, we extract feature values ​​for the forecast target through two-step analysis: “economic situation vector” and “economic forecast vector” (*2).

Step3 Prediction based on the extracted features

Using the feature values ​​extracted in Step 2 and the statistical data of 30,000 indexes as training data, we developed five prediction models and calculated the prediction values ​​for each. The final forecast results are calculated by ensemble considering the characteristics of each forecast model so that the overall forecast accuracy is optimized in various situations.
1 Structured data: Here, the contents of free-form Japanese sentences are analyzed, and data such as “automobile (Item) demand (Element) increase (Predicate) ⇒ automobile parts (Item) demand (Element) increase (Predicate)” is generated. Second, it decomposes the factors and results of events into individual elements and connects them as causal relationships.
2Economic situation vector, economic forecast vector: Proprietary patent-pending technology that quantifies when, how much, and in what direction the desired index will be affected when the content of a news article occurs. thing.

RELATED ARTICLES

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Recent Posts

Most Popular

Recent Comments