Jakob Bickley

This project examines how Alberta's generation mix has changed following the coal phase-out, with a focus on the rapid growth of wind, solar, and battery storage. Using AESO capacity and connection-queue data, it tracks how the grid's fuel composition has shifted and models what the pipeline of proposed connections means for supply adequacy and price risk going forward.

A central finding is the scale of proposed new load, including substantial data-centre demand driven by AI infrastructure expansion, relative to what the system can realistically absorb in the near term. The analysis frames this as a risk management question: grid operators and long-term energy buyers face real uncertainty around curtailment, pricing volatility, and firm supply adequacy as intermittent capacity grows faster than dispatchable alternatives.

• Tracked decade-long capacity trends by fuel type using AESO historical data and ggplot2 visualizations.
• Analyzed the connection-project queue to estimate near-term supply additions and their intermittency profile.
• Identified proposed data-centre load as a structural inflection point for Alberta grid planning and rate exposure.

Collaborative academic research project.

We built a Markov model to represent how bank withdrawals and asset liquidations evolve under changing interest rates and depositor sentiment. The model combines three components: a jump-augmented Vasicek process for interest rates (capturing both gradual drift and sudden rate shocks), Poisson processes for deposit inflows and outflows whose intensities shift with both the rate environment and prevailing confidence, and a three-state sentiment chain (Calm, Stressed, Panic) governing how quickly funding stress can escalate into a run.

We framed the problem as the risk management team at Silicon Valley Bank just before its 2023 collapse, calibrating the model's balance sheet parameters to SVB's actual structure. Across 100,000 Monte Carlo runs, the model estimated a mean loss of $0.38B for failed scenarios against SVB's true ~$1B realized loss, well within one standard deviation. The simulation makes clear how small sentiment shocks compound through fire-sale dynamics, and how a highly rate-sensitive depositor base is the primary driver of failure probability.

• Interest rates modeled via jump-augmented Vasicek SDE; simulated using a modified birth-death process (N = 10,000) interpolated to a 366-day daily grid.
• Confidence-state transitions (Calm → Stressed → Panic) driven by large withdrawal events and fire-sale signals; three asset classes: cash-like, short-term, and long-term.
• Failure time median 118 days (mean 138); mean conditional loss $0.38B, unconditional average $0.37B across all runs.