Rand Kwong Yew Low

Effective prioritisation of forested protected areas requires landscape-specific measures of relative intactness. Existing global intactness products rely on rule-based thresholds that may not be locally relevant, and can become obsolete in areas of rapid land-use change. We present a machine learning framework that predicts actual and alternative aboveground biomass density from spaceborne lidar footprints, from which disturbance, restoration potential and locally calibrated relative intactness can be derived. Static environmental features such as topography are held constant, while scenario features representing land cover and land use are modified at the prediction stage, isolating the effects of change. We provide a demonstration for the Taman Negara forest complex of Peninsular Malaysia, where outputs informed the delineation of Malaysia’s first tiger reserve. An estimated 60% of the landscape’s biomass shortfall predated the earliest satellite data. Intact old-growth forest in Phase 1 (924 km2) of the new protected area was distinguished from disturbed forest with high restoration potential in Phase 2 (416 km2). An alternative scenario for the construction of a road was used to calculate the loss of 152,869 ± 10,342 tonnes of old-growth aboveground biomass prevented by the gazettement of Phase 1. Phase 2 is due to be protected by 2028, and annual scenarios found no evidence of additional disturbance from anticipatory leakage. The results illustrate how alternative scenario predictions can support protected area delineation, restoration prioritisation, carbon accounting and annual monitoring. The framework can be transferred to and optimised for other landscapes in the tropical moist forest biome.

Market regime labels are not resolution-invariant: the same episode can be classified as ``crisis'' at one temporal resolution and ``calm'' at another. We fit Gaussian mixture models on log-volatility at four resolutions (5-minute to daily) for three assets across two geopolitical episodes, the 2026 US--Iran escalation and the 2022 Russia--Ukraine conflict, and measure cross-frequency agreement with the Adjusted Rand Index (ARI). For SPY, full-sample ARI is only 0.14, with the 15m--1h pair dropping to 0.07 despite well-separated fits. This severe dissonance replicates across assets and out-of-sample episodes. Aligning all frequencies to a common rolling window lifts intraday ARI to $\approx$0.48, so roughly half of intraday dissonance reflects window-length differences; the remaining half, and the entire daily--intraday gap, reflects genuine aggregation loss. The pattern is persistent, asset-specific, and survives alternative model choices, expanding-window estimation without look-ahead, and block-permutation tests that preserve regime autocorrelation. On 24 January 2026, all three assets exhibit mutually opposite regime assignments across resolutions. Regime-based risk frameworks consequently harbour a hidden temporal model risk that single-resolution backtesting cannot detect, underscoring the need for multi-scale monitoring in capital allocation and risk limit setting.

Retail investor participation in corporate bond markets remains limited, despite increased access via fintech platforms and rising demand for fixed-income products among aging populations. We examine why individuals stay disengaged through a structured multi-stage behavioural study. Using concept mapping, exploratory factor analysis (EFA), and confirmatory composite analysis (CCA), we identify key psychological and emotional drivers of investor participation. Across three sequential studies, trust, simplicity, brand familiarity, and perceived control consistently outweigh traditional financial metrics, such as yield or risk. These findings suggest that cognitive and emotional frameworks have a greater influence on investment decisions than technical understanding. The study also highlights concept mapping as a powerful tool for identifying behavioural drivers in fixed income markets. Our findings provide practical insights to strengthen personal finance decision-making and promote retail investor engagement by emphasising emotional trust, simplicity, and relatable product framing.

As Basel IV (the revised international regulatory capital framework) implementation proceeds, structural stability in regime models has evolved into a critical model risk governance mandate. We investigate representation uncertainty in market regime identification across four single-asset case studies spanning equities, sovereign debt, safe-haven commodities, and digital assets, utilizing multiple admissible rolling risk-based feature sets alongside Gaussian mixture and hidden Markov models. Across these asset classes, the cross-representation agreement is persistently low and remains materially inferior to temporal agreement under fixed specifications. A step size sweep uncovers a crucial artifact: much of the apparent temporal stability is mechanically inflated by rolling-window overlap, whereas representation-driven disagreement persists regardless of step size. Consequently, partition-level regime assignments exhibit minimal evidence of representation invariance within the economically admissible classes studied here. However, the evidence is not of wholesale instability; we find that coarse risk ordering remains substantially more robust than exact partition boundaries. For governance, the implication is direct: the uncertainty of representation must be recognized as a first-order source of model risk. We argue that cross-representation diagnostics should become a standard complement to conventional temporal validation in risk-sensitive financial applications.

We analyze the dynamic evolution of volatility transmission both in strength and direction among constituent sectors of the Australian Stock Exchange (ASX) from 2005 to 2023, spanning major disruptions such as the Global Financial Crisis (GFC), oil price crash, implementation of carbon tax policy, and the COVID-19 pandemic. We construct the spillover indices using the Time-Varying Parameter (TVP)-Vector Autoregression (VAR)-Stochastic Volatility (SV) model combined with the DY spillover index. Network graphs highlight persistent volatility spillover trends and structural shifts across crises. During the GFC, the highly leveraged REIT sector transmitted volatility to the Financials and Industrial sectors, exposing credit market fragility. In contrast, the pandemic-induced lockdown and transition to remote work saw REITs become major volatility receivers from IT. The carbon tax policy induced lasting uncertainty in the Energy sector with effects persisting even after the policy’s withdrawal. During the pandemic, the discretionary sector became a more dominant receiver amid increasing stresses from Healthcare, IT and Utilities sectors. Our findings provide critical insights to investors and policymakers to account for evolving sectoral linkages when mitigating market stress.

We examine whether news sentiment predicts corporate bond returns in both the cross-section and time series. Using univariate and double-sorted portfolios by investment grade, issuer size, maturity, and liquidity, we find that while some results are statistically significant, the economic magnitudes are negligible. Unlike evidence in equities and the limited bond literature, our findings suggest that news sentiment does not meaningfully explain the cross-sectional variation in corporate bond returns and provides no incremental pricing power as a bond risk factor.

We discuss three applications of blockchain data technology that illustrate its considerable problem-solving potential in: (i) Centralized Bank Digital Currencies (CBDC); (ii) Healthcare (HC); and (iii) Non-Profit Organizations (NPOs). Key solution features include security and immutability, along with authentication in a decentralized network that can yield the same consensus solution as a single centralized computer would. But notwithstanding the strength of blockchain’s security, vulnerabilities in the wider infrastructure of the applications we considered. We discuss real-world vulnerabilities in error correction and smart contract code, and the integration of blockchain data and infrastructure that is essential in day-to-day operation. Further, the decentralization in this (Web 2.0) network infrastructure is, if not the proverbial “bug”, a weakness and decidedly not a feature.

Blockchain technology can enable non-profit organisations (NPOs) to create a large, diverse, and engaged community of donors and stakeholders in an autonomous and transparent manner