Debashis Chatterjee

Purchasing Managers' Index (PMI) surveys provide high-frequency, firm-level information about economic conditions. Beyond the headline PMI, the surveys report conceptually distinct components: forward-looking expectations (Future Output), contemporaneous quantities (Output and New Orders), labor demand (Employment), and price pressures (Input Prices and Output Prices). This paper asks which type of PMI information contains the most economically meaningful predictive content for the future path of GDP growth and to what extent forward-looking expectations can be used to construct accurate forecasted GDP values relative to other contemporaneous indicators. Using the provided quarterly dataset (2012Q2-2022Q1), we align PMI information observed in quarter t with GDP growth in quarter t+1 and evaluate predictive content through: (i) predictive regressions with heteroskedasticity-and-autocorrelation-consistent (HAC) inference, (ii) pseudo out-of-sample forecasting with expanding-window estimation, and (iii) a Shapley-value decomposition of regression R 2 to quantify each component's marginal contribution under multicollinearity. The methodology is designed to clarify the micro-to-macro information channel and rigorously compare expectations against realised quantities, labor, and prices.

Mechanistic compartmental models such as SIR provide interpretability and enforce conservation of population, but their predictive performance can degrade in heterogeneous settings where transmission varies over time and across regions. Conversely, spatiotemporal graph neural networks (GNNs) can learn flexible spatial interactions from data, yet unconstrained predictors may yield epidemiologically inconsistent dynamics. We propose MechGNN-Epi, a hybrid framework that couples a spatiotemporal graph encoder with a differentiable SIR update. Rather than directly predicting future infections, the model learns region- and time-specific transmission and recovery parameters from graph-based latent representations and then propagates compartments through a mechanistic solver. This design yields epidemiologically constrained trajectories and produces region- and time-indexed parameter proxies that can be inspected as diagnostic signals, while not being guaranteed as causally identifiable mechanistic parameters. We evaluate the pipeline on U.S. state/territory COVID-19 surveillance data using strict chronological splits and report variability across multiple random seeds, multi-horizon rollouts, and ablation/sensitivity studies against graph and model variants.

The Fiji–Tonga–Kermadec convergent margin represents one of the most seismically volatile regions globally, defined by complex subduction dynamics and a high density of deep-focus events. This study performs a high-resolution exploratory investigation of a 1000-event seismic catalog to resolve the intricate interplay between rupture magnitude, focal depth, and spatial distribution. Moving beyond baseline descriptive summaries, the study employs Exploratory Spatial Data Analysis (ESDA), including non-parametric 2D kernel density estimation and hexagonal binning, to delineate the latent tectonic architecture of the region. Our analysis identifies a pronounced bimodal depth signature, with focal concentrations in the shallow crust (500 km). These findings empirically validate the thermal and phase-transformation mechanics of the Wadati–Benioffzone. Multivariate diagnostics further reveal a robust linear dependency 𝑟 = 0.71 between seismic magnitude and the global monitoring footprint, providing a quantitative metric for energy-dependent detectability across the station network. By bridging classical geostatistical workflows with modern spatial intensity modeling, this research provides a reproducible, data-driven foundation for enhancing seismic hazard zoning. These insights are particularly critical for improving infrastructure resilience and disaster risk reduction (DRR) strategies in Small Island Developing States (SIDS) facing cascading tectonic risks.

Cities shape mosquito and arboviral risk through the way heat, water, vegetation, and drainage are managed. We present a fully reproducible early-warning pipeline that predicts trap–week West Nile virus (WNV) positivity one to two weeks ahead using programmatically accessible sources (Socrata WNV pools and Meteostat weather), and a graph-aware post-hoc smoothing that enforces spatial coherence of predicted risks. Because direct API access to the Chicago portal requires a Socrata App Token, all reported experiments use an offline, synthetic trap–week panel that mirrors a Chicago-like spatio-seasonal structure, while we ship turn-key code to re-run the identical analysis on the real data once a token is provided. Across forward-chaining evaluations, gradient boosting with Laplacian smoothing delivers discrimination comparable to that of a strong elastic-net baseline, as assessed by the area under the receiver operating characteristic curve, but substantially better probabilistic calibration (lower Brier and improved reliability). Feature profiles emphasize antecedent heat and relative dryness, aligning with ecological priors, and enable tiered, probability-based operational playbooks for vector control and sustainability co-actions (cooling, drainage, vegetation). The pipeline is designed for transparency, portability, and policy relevance: calibrated probabilities support graded interventions and top-K targeting under budgets, while code parity between synthetic and real modes facilitates external replication.

We propose the AI-modulated Pólya Urn(AIM-PU), a stochastic framework bridging classical path-dependent reinforcement processes with adaptive, covariate-driven decision rules. While standard Pólya--Eggenberger urns assume fixed replacement kernels, AIM-PU introduces a heterogeneous reinforcement mechanism where the replacement tensor is modulated by a learned policy adapted to a covariate filtration. We establish that this coupling preserves analytical tractability via martingale approximations. Under mild assumptions on policy regularity and graph irreducibility, we prove: (i) controlled linear growth of the total mass; (ii) almost-sure convergence of the normalized composition; and (iii) a functional central limit theorem (FCLT) where fluctuations converge to a Gaussian mixture. The framework is first verified through extensive synthetic simulation studies that confirm the theoretical convergence rates and distributional stability in volatile, non-stationary environments. We then demonstrate the model's practical utility in risk-sensitive portfolio allocation using real-world financial data (SPY, MSFT, DUK, TSLA) from 2020--2025. By embedding Conditional Value at Risk (CVaR) constraints directly into the urn dynamics, AIM-PU structurally enforces long-term risk limits. Our results show that AIM-PU outperforms stateless contextual bandits in both Sharpe ratio and drawdown minimization, providing a robust, interpretable, and mathematically grounded approach to automated financial decision-making.

We introduce the congestion–discouraged Poisson (CDP) distribution, a two‐parameter count model derived from a state‐dependent birth–death process where arrivals are suppressed by system backlog. The distribution is specifically designed to model underdispersed count data common in operations research, such as service queues and inventory pipelines. We establish its fundamental properties, including ultra‐log‐concavity, a Stein‐type identity, and a simple probability recursion. Likelihood‐based inference is developed, with parameters providing direct operational insights into congestion costs and idle probabilities. In a simulation study and an empirical application to the WomenQueue dataset, the CDP model outperformed both Poisson and Conway–Maxwell–Poisson (CMP) benchmarks in AIC, BIC, and Pearson X ² .The results confirm that the CDP distribution is a principled mechanism‐based tool that accurately captures congestion‐sensitive dynamics while remaining as tractable as classical benchmarks.