LinkedIn Series
All posts in one place. Read in order or jump to what you need.
Overview & How to Use
This page collects my LinkedIn series in a playlist-style format. Each series has a short description and a numbered list of posts with direct links to LinkedIn. Start at Week 1 to follow the full narrative, or skim the summaries and jump into a topic that matches your needs.
Experimentation in Airlines
PlaylistWeek 1 · Why airline experiments are hard
November 19, 2025
The post outlines why airline pricing experiments are especially difficult. Pricing affects demand, competitors, and capacity, so flights are not independent. Booking data are autocorrelated and noisy, which makes detecting small effects hard. Despite this, experimentation is essential because it provides clean causal evidence and even small revenue gains matter at scale. Building this capability strengthens pricing decisions and lays the foundation for broader experimentation across the airline.
Week 2 · When A/B testing meets airline reality
November 26, 2025
The post shows why standard A/B testing breaks in airlines. Treatment and control cannot stay independent because customers share the same seats, routes influence each other, competitors react, and booking curves create timing effects. To get reliable results, airlines use designs that limit these spillovers, such as route-level randomization, pods, switchbacks, and models that account for network effects.
State Space Models
PlaylistWeek 1 · What is a State Space Model?
August 13, 2025
Introduces state space models as a way to separate signal from noise, combining interpretable components (trend, seasonalities, cycles, regressors) with Kalman filtering for estimation.
Week 2 · Why use State Space Models instead of ARIMA?
August 20, 2025
Compares SSMs with ARIMA, highlighting SSM strengths for trends, breaks, and missing values and notes that ARIMA itself can be written in state space form.
Week 3 · The Building Blocks: Trend, Seasonality, Cycle, Irregulars
August 27, 2025
Shows how SSMs are built like Lego: combine trend, seasonalities, cycles, and regressors to decompose and explain complex time series.
Week 4 · The Art of Optimal Updating: The Kalman Filter
September 3, 2025
Explains the Kalman filter’s prediction–update logic and why it yields optimal state estimates given uncertainty in data and model.
Week 5 · How Hindsight Improves Estimation: The Kalman Smoother
September 10, 2025
Shows how the smoother uses the full dataset to refine past state estimates, revealing clearer trends and breaks.
Week 6 · Missing Data? No Problem: The Kalman Filter
September 17, 2025
Demonstrates how the filter naturally handles gaps by propagating the model forward, with uncertainty widening until data resumes.
Week 7 · Forecasting with State Space Models
September 24, 2025
Covers multi-step forecasting by treating future points as missing, inspecting component-wise forecasts and uncertainty.
Week 8 · From regression to state space: adding covariates the smart way
October 1, 2025
Explains including covariates as deterministic regressors (variance set to zero) so betas are interpretable with standard errors equal to recursive least squares.
Week 9 · Predicted, Filtered, and Smoothed Errors
October 8, 2025
Defines the three residual types and why one-step prediction errors are the right diagnostic for model adequacy (ACF flatness, zero mean, homoskedasticity).
Week 10 · From prediction errors to learning: how the Kalman filter uses likelihood
October 15, 2025
Connects prediction errors and their variance to the likelihood; shows how maximizing likelihood tunes parameters (e.g., signal-to-noise ratio) for the best fit.
Week 11 · The Kalman Filter’s Linearity Assumption
October 29, 2025
The Kalman filter’s closed-form beauty relies on linear and Gaussian assumptions; the Oxford–Cambridge Boat Race reveals what happens when those assumptions break.
Missing a post? Let me know and I’ll add it.