Artificial intelligence has been reshaping vast parts of the economy, and the financial sector is no different. The use of algorithms for stock picking has been written about considerably. Its use for trading in debt markets, however, is perhaps more interesting. When applied to bond trading, the implications move beyond the financial and into the political, with government financing made simultaneously easier and more vulnerable.
Unlike traditional bond trading, algorithmic trading uses a computer program that follows a defined set of instructions to place trades. These are intended to use data at a frequency and speed beyond human abilities.
The advantages of algorithmic trading are clear enough. Algorithms can deploy market scanning at a speed and scale that humans cannot. This leads to the early identification of market opportunities and quick detection of price differences, enabling successful arbitrage against less sophisticated market participants
Another clear advantage is risk management. Advanced algorithms can have in-built risk management features which follow predefined risk parameters and adjust trading activity accordingly. These are ideal for more conservative trading strategies.
But within the debate on algorithmic trading, its impact on the crucial bond markets are worth exploring.
Rotem Farkash: AI Deep Learning allows for the incorporation of more information beyond narrow technical factors
Algorithms are greatly customizable, meaning that traders can optimize them to fit their particular trading objectives and style. With the addition of artificial intelligence features such as machine learning, these customized algorithms can improve by absorbing historical data and market feedback.
Natural Language Processing (NLP) a subfield of AI that trains machines to understand human language, also means that AI is becoming increasingly competent at interpreting qualitative information from the news, social media, and earnings reports, allowing it to anticipate market movement based on human sentiment.
Tech entrepreneur and AI expert Rotem Farkash has explained that the possibilities for machine learning in bond finance are just beginning. ‘AI Deep learning allows for the algorithm to move beyond narrow technical factors, and over time it will only improve through its in-built feedback mechanisms’, Farkash explained.
‘Tools will be able to assess not just statistical variables on bonds like yield and real return, but qualitative factors like the current government’s fiscal policy,’ Farkash continued.
European Central Bank: supply and demand for financial assets may be distorted systematically
In bond markets, as in equity markets, algorithms and artificial intelligence mean less human error-based volatility and greater speed of trading, pricing and arbitrage. This enhanced speed and information flow makes markets more efficient and can avoid certain kinds of irrational activity.
Despite this, algorithms bring their own risks. If many firms all use AI algorithms from the same providers for their trading strategies, then they will all follow the same actions at the same time. In certain scenarios this can mean sudden swings in the market based on new information, creating the potential for new crashes and greater systemic vulnerabilities.
As the European Central Bank has cautioned ‘should many institutions use AI for asset allocation and rely only on a few AI providers, […] then supply and demand for financial assets may be distorted systematically, triggering costly adjustments in markets that harm their resilience’. This is also before the possibility of ‘Black Swan’ events not reflected in historic data, is taken into account.
Greater demand from retail investors using AI will push up bond prices and lower borrowing costs but markets will become more sensitive
The biggest fundamental change is that more buyers will enter the bond market which will make borrowing cheaper for governments. However, these new buyers will be using algorithmic strategies that are more sensitive and faster than traditional human traders. This ultimately means markets are deeper and more liquid but also more sensitive and potentially more aggressive.
Increased capital from the perspective of the state is an unalloyed good. Greater demand from retail investors will push up bond prices and in turn lower borrowing costs, allowing governments to manage increasingly strained budgets.
The greater arbitrage enabled by these techniques, however, can equally become a problem for governments. The heightened sensitivity of algorithm-led investors means governments will have work harder to keep their bondholders and buyers happy, which will constrain their freedom in economic policy. Debates about the excessive power and influence of debt traders on sovereign democratic government will only intensify in such a scenario.
The data weaknesses and biases of algorithms will need human oversight
Inevitably, the greater liquidity that algorithmic trading brings will be welcomed by traders and governments alike. Its volatility risks, however, will have to be managed by human oversight, just as AI has been combined with human supervision in other sectors.
Until Natural Language Processing overcomes the inability of AI to make qualitative and contextual judgements, bond trading will need a human oversight to keep it stable. In the meantime, bond markets will enjoy greater demand but also enhanced algorithmic sensitivity. For governments, they will have to account for more reactive markets, the dangers of sudden algorithmic swings and any biases in these tools from incomplete data.
