---

Mizuho’s new AI system to make the electronic trading of government bonds faster and more accurate.

 

Traders are no longer reliant on intuition and experience as AI transforms the world of government bond trading

Recently, artificial intelligence (AI) has been making its presence felt across numerous, diverse industries. In the financial sector, this has led to the rise of new products and services such as AI-powered investment and trading. Mizuho continues to develop and deploy AI-driven tools to make this transformational technology of practical use to investors. One of the latest examples is “ALPHA,” a new system developed by Mizuho Securities that utilizes AI for the electronic trading of government bonds. One major advantage of this system is the way it gives traders real-time analytics and advice to help them offer better rates to customers. Until now, traders had to produce the most-accurate rate possible for each individual order. In a trading environment where every second counts, traders can find themselves in a tight spot when several large orders come in at the same time. In the past, each trader also had to rely on qualitative factors, such as intuition or experience. ALPHA has been developed to provide AI support and to offer multiple rates in a swifter, quantitative, and more-accurate manner. Amid growing interest in the substantial benefits of utilizing AI in investment and trading, this new, innovative combination of government bonds and AI is making huge waves in the industry.

ALPHA analyzes 215 types of data in real time to generate the best rate

In 2019, in response to an increase in the electronic trading of government bonds, Mizuho began using “Tangoro,” a system that automatically offers rates for certain individual orders. This system enabled Mizuho to handle more orders quicker, but we saw room for improvement when it came to quality. Namely, how to provide customers with even better rates and optimize post-contract hedges. ALPHA was developed to improve these attributes.

With ALPHA, AI analyzes past trades and current market conditions in order to provide traders with intelligence about the market impact of each trade and correlated hedging methods. Within this scheme, ALPHA synthesizes data in real time, each time a customer order arrives.

For example, when a customer provides information about the type and number of securities they wish to purchase, ALPHA will analyze 215 categories of data in real time, including the inquiring customer’s past trades, events pertaining to central bank purchasing operations or JGB auctions, interest rates, and individual companies’ risk values. The AI will then use machine learning to predict how the market will change in the immediate term.

Based on these predictions, ALPHA will provide traders with two pieces of advice. The first, is whether the trader, based on constantly-shifting market data, can improve the rate offered for the trade in question. The second, is selecting which kind of hedge should be used to mitigate risk.

Traders once had to ponder the variables themselves, with little help, but ALPHA can now do the thinking for them. Furthermore, ALPHA can provide a recommendation approximately 0.2 seconds after the customer inquiry is received. The trader can decide whether ALPHA’s advice is applicable to the customer’s inquiry, but ALPHA is ready to help when several orders flood in at once and the trader has to act instantaneously, or when the trade is complex and the trader has doubts about their ability to make the right call. ALPHA, in other words, is a partner that is there to support the trader.

From planning to release in just nine months – a system developed entirely in-house

ALPHA has been developed and managed entirely in-house at Mizuho.  At the conceptual stage, we had doubts about the feasibility of achieving our desired goals. But outsourcing seemed difficult, because the system specifications would need to change frequently, in tandem with market shifts and traders’ wishes.

After deciding to handle the entire development internally, we convened professionals from several fields, including experts in program development, systems management, and the construction of machine learning models.

Of course, the system had to be user friendly, and ALPHA had to be incorporated seamlessly with the workflow of traders involved in the electronic trading of government bonds.

Several Mizuho traders participated in the development, sharing their thoughts on all aspects of production both big and small, from planning, data analysis, model preparation, program development, and server management, right down to how the user interface should appear.

The ALPHA project launched in April 2021. From there, things moved quickly – the team was fully assembled by July, and operations commenced in December. The swift nine-month cycle, from planning to release, was a result of the project being developed entirely in-house, and by a diverse pool of talent.

In an age of uncertainties, we must never stop thinking

ALPHA continued to run smoothly after its release, but the usually-tranquil government bond market was soon rocked by the Russia-Ukraine crisis and global inflation. Of course, ALPHA’s learning had not incorporated these unique scenarios, so new rules were added to the system. These added improvements help ensure ALPHA does not give risky advice when unexpected events occur, and that it “would not do anything a trader wouldn’t do” when faced with such situations.

Furthermore, the AI model is renewed approximately once every two weeks to ensure that ALPHA’s learning is as up-to-date as possible. This ability to continually update the system to respond swiftly to ever-changing events is another major benefit of developing the system in house.
 

At Mizuho, we strongly believe that if the evolution of finance requires something that does not yet exist, we just have to invent it ourselves!

ALPHA epitomizes this can-do attitude. As we strive toward a brighter financial future, Mizuho will continue to create new value based on our spirit of continued improvement and the strength of the people and teams that embody this spirit.

 

Text and photographs by Mizuho’s DX Editorial Team

 

---