Investor Spotlight: KWAP - navigating technological shifts in quant equities

Mohd Ridzuan Mohd Nasir
Vice President, Investment Excellence Function, KWAP

Kumpulan Wang Persaraan (Diperbadankan) (KWAP) is Malaysia’s public sector pension fund and a key government-linked investment institution. Over the past decade it has built a quantitative equity programme that began as a research-led pilot allocation and has matured into a live, in-house mandate within the fund’s domestic equity portfolio. That experience is now informing a fresh chapter. Through its InnoLab platform, KWAP is evaluating opportunities in long-only systematic global equity strategies that incorporate artificial intelligence (AI) techniques.

These new strategies are subject to a structured evaluation process, with clear governance, risk controls, and ongoing oversight prior to any broader portfolio integration, reflecting KWAP’s responsibility as a public sector pension fund. Mohd Ridzuan Mohd Nasir, who heads InnoLab and has been at the centre of KWAP’s quantitative journey, spoke to bfinance Investor Spotlight about the organisation’s experience and insights from its quantitative equity journey.

What is driving KWAP’s exploration of AI-integrated strategies?

KWAP has gained valuable experience with multi-factor quantitative equities, through both internally managed domestic Malaysian equities and externally managed global equities. However, the data processing and decision-making frameworks of AI-driven strategies differ materially, with AI serving a more prominent role in the stock selection process. These strategies leverage large datasets and aim to capture complex, non-linear relationships that traditional quantitative approaches may not fully identify. This is what is driving KWAP’s interest in the space, assessing whether AI-driven techniques can complement our existing quantitative capabilities and provide an additional source of diversified risk-adjusted returns.

What is your thesis for AI-driven quantitative equity investing?

Quantitative strategies typically follow a structured and disciplined process which includes generating signals, constructing portfolios, and executing trades. Traditional approaches often rely on more linear relationships, which may not fully capture complex, non-linear drivers of return. Incorporating non-linear analysis can help uncover additional sources of diversified risk-adjusted returns. These techniques span a spectrum, from established discriminative tree-based models such as XGBoost and LightGBM, which typically remain more prevalent in investing due to their stability and consistency, to generative natural language processing (NLP)-driven neural network and transformer models that can extract insights from unstructured data. In addition, alternative data such as sentiment, news flow, and earnings call transcripts may provide further dimensions in identifying return opportunities. When considered holistically alongside traditional data, these inputs can enhance the predictive power of expected returns and stock rankings. In addition to this, the level of data complexity has evolved considerably since we began factor investing more than a decade ago. Our thesis is that AI-driven techniques can extract signals from new, less easily processed data sources, and complement existing return drivers within the wider portfolio.

Going back to the start, what originally motivated KWAP to explore quantitative approaches to equities?

KWAP needed to find new, diversified sources of return, pursued through a much more disciplined and systematic approach than was typical at the time. Factor investing, which generally relies on a more linear methodology when assessing the relationship between fundamental data and expected returns, had been in the academic literature for many decades, but it was still relatively new for many Malaysian institutional investors, and to some extent remains so today. At the time, KWAP was among the early adopters within Malaysia’s government-linked investment institutions implementing factor-based portfolio strategies. From both a quantitative and investment perspective, we observed early on the coverage biases of purely fundamental analysis, the behavioural biases that creep into stock evaluation, and the dominance of narrative. There was a clear need to look at the whole data set and apply an unbiased, statistical approach. We presented ideas internally for about a year, but it became clear that the work would only gain traction through disciplined implementation and actual portfolio deployment under a systematic approach.

KWAP began with domestic Malaysian equities rather than global markets. What was the rationale for starting there?

Initially the natural inclination was to go global, given the breadth of data available. But we also wanted to wear the national hat and prove to ourselves, and to external managers, that systematic models could be implemented in a smaller market such as Malaysia. As a government-linked investment institution, we also saw it as part of our stewardship role to help foster institutional quantitative investing capabilities within the domestic market. Breadth is more limited than in the MSCI ACWI, but data is data. So, we asked whether we could still capture the return opportunities across a narrower universe. That meant tackling liquidity and turnover constraints head on. Starting domestically also allowed us to establish the partnership with our relevant external fund managers to run the global multi-factor strategies in parallel, which provided differentiated quantitative insights and model governance to support our in-house portfolio.

What were the biggest gaps between academic theory and live implementation?

Academic backtesting tends to assume frictionless execution. The moment you manage a live quantitative investment portfolio, transaction costs, market impact, turnover and volume constraints become first-order considerations, particularly in a smaller market. Trades that an academic model completes almost instantaneously may, in practice, need to be spread across several days to preserve the expected returns, though not to the extent that execution becomes overly prolonged, as this can introduce opportunity costs and dilute the intended return expectations. We worked closely with our central dealing team to become more price-sensitive on execution, which itself became an additional source of excess return on top of the quant model. The other insight was that liquidity itself can be a source of return. Rather than excluding less liquid stocks, we explored whether a liquidity premium could be harvested, provided governance and risk controls were sufficient. Covid-19 was the most important live test. No backtest can anticipate that kind of speed or severity of dislocation, but the model proved resilient because we did not override it. Discipline matters above all.

As part of engaging external global managers, what did ‘knowledge transfer’ actually involve?

Knowledge transfer is only superficial if you allow it to be. We were never trying to replicate the external managers’ intellectual property (IP). The engagement was about validating our own IP and the model from a governance and risk-management perspective, and ensuring our approach was consistent with industry best practice. That happened through active engagement on methodology, comparing notes as we managed the portfolios in parallel, and a structured approach we called ‘LEAP’ (Learning Engagement & Advisory Programme). Human capital development was as much a priority as investment capability development. As newer techniques such as NLP began to feature in the managers’ work, those conversations also helped shape our view on whether AI should be applied primarily as a tool to automate existing tasks, or as a core component of an investment strategy.

How did you bring the wider organisation along with you?

KWAP management was very supportive from the outset. The bigger task was explaining to stakeholders how the model worked and how risks were being managed. Systematic strategies can feel less intuitive to colleagues accustomed to a fundamental approach. Once we could articulate the governance and nuances around the model, the buy-in followed, further strengthened by the credibility gained from developing our own in-house quantitative model from the ground up. The fact that the model performed well alongside our fundamental colleagues, often across different market environments, made the case for diversification of style within the equity allocation.

What design choices have made the model resilient rather than fragile?

The decisive choice was to ensure fully diversified exposure across multiple factors and to avoid concentration in any one of them. Some managers hold a philosophical conviction on fixed factor weights. Our approach is to optimise those weights dynamically, so that the portfolio can adapt to changing market regimes while still capturing short-term opportunities. The other discipline is to accept that most short-term news flow may be noise, which systematic strategies can look through, a competitive advantage in itself.

As AI-driven strategies become more prominent, what should asset owners be asking to distinguish genuine innovation from repackaged factor models?

Be selective and look closely at methodology. In our engagements with external managers, we were particularly focused on distinguishing genuinely differentiated approaches from traditional factor models presented under a new AI label. The objective was to assess whether managers were using large and diverse datasets in ways that generate differentiated, risk-adjusted sources of return in comparison to traditional quants. While many techniques now grouped under AI build on long-established statistical and computational methods, advances in computing power and data availability have encouraged their adoption and made them more accessible and investable. Ultimately, technology must be supported by human judgement, oversight and governance. AI and human intelligence should work together, not separately.

What is the most important lesson for peers contemplating the next wave of systematic or AI-driven equity strategies?

Deploy within a controlled, well-governed environment, at a deliberately small and manageable scale, before considering any broader portfolio integration. We don’t see innovation as ‘aiming to fail’ but creating a disciplined environment that allows for learning and refinement; this definition sits well for a public pension fund, where we have a duty of stewardship that cannot be compromised. Equally, it is important to recognise the role of human capital in this context. The most valuable thing we have built is a dedicated in-house team with the skills to evaluate models, engage with external partners on equal terms, and develop their own thinking over time. Technology evolves quickly, but the capability to assess and apply it effectively is what endures.