Financial markets increasingly depend on data, algorithms and quantitative analysis. Yet behind every model is a person deciding which information matters, which assumptions are reasonable and when an apparently strong result deserves closer examination.
Jason Lam joined Global Investment Strategy Hong Kong Limited in 2024 as a Quantitative Trader. He holds a Master’s degree in Mathematics from the Chinese University of Hong Kong, having previously studied Aerospace Engineering.
At GIS Hong Kong, Jason uses quantitative techniques in research, data modelling, trading analysis and risk monitoring. We spoke to him about his unconventional route into financial markets, the practical use of machine learning and why more complicated models do not necessarily produce better answers.
You originally studied Aerospace Engineering. What led you from engineering into mathematics and quantitative trading?
Aerospace Engineering gave me a strong grounding in applied mathematics, modelling and problem-solving. You learn to take a complex system, identify the forces acting upon it and consider how a change in one variable might affect the wider structure.
Financial markets are clearly different from aircraft, but some of the underlying disciplines transfer surprisingly well. Both involve changing systems, incomplete information, and outcomes that cannot be understood by looking at a single variable in isolation.
I became increasingly interested in the mathematical side of those problems. That led me to study Mathematics at Master’s level and eventually apply that knowledge within quantitative trading.
How does an engineering mindset influence the way you develop a trading strategy?
Engineering encourages a practical mindset. A model must do more than produce an elegant result; it has to work under realistic conditions.
A systematic strategy usually begins with an idea or observed relationship that can be expressed and tested objectively. We then ask whether it is persistent, economically meaningful and robust across different periods and market conditions. Realistic implementation costs and potential failure points also have to be considered.
In the broader context, a convincing backtest is only the beginning. You still need to understand why the relationship might exist and what could cause it to break down. If there is no credible rationale behind the result, it may simply be a pattern that happened to appear in the historical data.
How do you distinguish a meaningful signal from something that has appeared by chance?
Financial datasets contain an enormous number of relationships, so some will appear significant purely by chance.
One danger is repeatedly changing a model until it explains the historical data extremely well. That may create the impression of precision, but it can make the model less likely to work when presented with new information.
We test signals using data that was not involved in their development and examine whether the result survives reasonable changes to the assumptions, periods and parameters. If a strategy only works under one very specific set of conditions, that is usually a warning rather than a strength.
You also have to ask whether the result makes economic sense. Statistical significance on its own does not explain why an opportunity exists or whether it is likely to persist.
Can quantitative analysis improve risk management without creating false confidence?
It allows us to examine volatility, concentration, correlations, potential drawdowns and performance under different scenarios. That provides a more structured picture than looking only at an expected return.
Multi-asset portfolios can become more difficult to manage because investments do not always behave in the same way under different market conditions. Shares and bonds may appear to provide diversification during normal periods but begin moving in the same direction when markets come under pressure.
Precision can also be misleading. Expected returns, volatility and correlations are not fixed, while relatively small changes in the inputs can produce large changes in an optimised portfolio.
Quantitative analysis helps identify exposures; it does not reduce risk to one unquestionable number. Different measures reveal different characteristics, and the wider market context still matters.
As machine learning becomes more capable, where does human judgement still matter?
Machine learning can process large datasets and identify relationships that may be difficult to detect using simpler methods. It can support areas such as signal research, market classification, portfolio construction and risk monitoring.
But greater complexity is not automatically an improvement. A complicated model may be harder to test and more difficult to understand when conditions change. If you cannot explain why it produced a result, it becomes harder to judge whether that result remains reliable.
People still decide which data to use, how a model is structured, which constraints apply and when its assumptions need to be reviewed.
Human judgement should not mean overriding a model whenever its conclusion feels uncomfortable. It means understanding the model well enough to recognise its limitations and determine whether it is still being used for the purpose for which it was designed.
What makes Hong Kong an interesting base for quantitative trading?
Hong Kong sits where regional and global influences meet. APAC markets differ significantly in liquidity, market structure, regulation and investor participation, creating both analytical challenges and opportunities.
Developments in mainland China, international monetary policy, currencies and global risk sentiment can affect markets across the region in different ways. A model developed for one market cannot simply be transferred to another without accounting for those differences.
That makes local market knowledge just as important as analytical capability. Data may reveal a relationship, but understanding the structure and participants of the relevant market provides the context needed to interpret it properly.
Which development will have the greatest practical effect on quantitative finance over the next few years?
Better data, greater computing power and improved machine-learning techniques will allow researchers to examine more information and test ideas more efficiently.
The more important change, however, will be how those capabilities are incorporated into actual trading and risk processes. A technically impressive model has limited value if it cannot be tested properly, explained to the people responsible for it or used within appropriate controls.
Markets will remain uncertain regardless of how sophisticated the technology becomes. The objective is not to eliminate that uncertainty, it is to understand it better and make more disciplined decisions within it.
What is the most important lesson you have learned from working with models?
Complexity can be impressive, but clarity is more useful.
If you cannot explain what a model is doing, which assumptions it depends upon and where it might fail, you should be cautious about relying on its answer.
Disclaimer: Users of this website should seek independent financial advice prior to making any investment decision and no information herein shall constitute general or specific investment advice of any kind.