I beseech you, in the bowels of Christ, think it possible you may be mistaken. - Oliver Cromwell

• A regular time series is a function from integers to real numbers: \(y_t = f(t)\).
• Many useful time series can be specified using linear difference equations like \(y_t = k_1y_{t-1} + k_2y_{t-2} + \dots + k_ny_{t-n}\)
• This recurrence relation has a characteristic equation (and matrix representation), whose roots (or matrix eigenvalues) can be used to write closed-form solutions like \(y_t=ax^t\).
• Any time series combining exponential growth/decay and sinusoidal components can be modeled by a linear difference equation or its matrix representation.

PCA is an essential data science tool which uses the SVD to break down the linear relationships in data. The Gavish-Donoho optimal truncation threshold provides a simple formula to select a good threshold for dimensionality reduction.

Meditating on practical open distributed computing, how to build un-take-down-able apps like Web3 but without permissionless blockchains.

The perpetual challenge is building upper tiers before lower tiers are 100%, and strengthening lower tiers without breaking upper tiers.

Revisiting Bengen’s “4% Rule” at various levels of risk aversion, and generalizing beyond a simple fixed-withdrawal, no-shortfall rule, to flexible rules at different levels of risk aversion.

“Our decision to permanently suspend Donald Trump from the Twitter platform, may be a major inflection point in Twitter’s history. As CEO, I owe our users and employees a clear statement of why we took this action and how this decision evolved, i.e. not just some pablum about what a hard decision and potentially dangerous decision it was.”

Do you want to do fast and easy portfolio optimization with Python? Then CVXOPT, and this post, are for you! Here’s a gentle intro to portfolio theory and some code to get you started.

Bayesian optimization of machine learning model hyperparameters works faster and better than grid search. Here’s how we can speed up hyperparameter tuning using 1) Bayesian optimization with Hyperopt and Optuna, running on… 2) the Ray distributed machine learning framework, with a unified API to many hyperparameter search algos and early stopping schedulers, and… 3) a distributed cluster of cloud instances for even faster tuning.

A while back I made this Pizza service weekend project and I thought I could just press a button in AWS and deploy it in the cloud. It turned out to be… more complicated. With the latest version of Docker it’s getting easier. Here’s the harder (old) way and the easier (new) way. After some configuration, you can just say docker compose up and your container is deployed.

A Docker configuration for machine learning

I loved The Biggest Bluff: How I Learned to Pay Attention, Master Myself, and Win, by Maria Konnikova.

Reinforcement learning is a machine learning paradigm that can learn behavior to achieve maximum reward in complex dynamic environments, as simple as Tic-Tac-Toe, or as complex as Go, and options trading. In this post, we will try to explain what reinforcement learning is, share code to apply it, and references to learn more about it. First, we’ll learn a simple algorithm to play Tic-Tac-Toe, then learn to trade a non-random price series. Finally, we’ll talk about how reinforcement learning can master complex financial concepts like option pricing and optimal diversification.

Shiller’s CAPE ratio is a popular and useful metric for measuring whether stock prices are overvalued or undervalued relative to earnings. Recently, Vanguard analysts Haifeng Wang, Harshdeep Singh Ahluwalia, Roger A. Aliaga-Díaz, and Joseph H. Davis have written a very interesting paper on forecasting equity returns using Shiller’s CAPE and machine learning: “The Best of Both Worlds: Forecasting US Equity Market Returns using a Hybrid Machine Learning – Time Series Approach“, which effectively applies machine learning to an import investing problem. Image source: Improving U.S. stock return forecasts: A “fair-value” CAPE approach, Joseph Davis, et al. (2017)

As a data scientist, you might say…“A blog post about thresholds? It’s not even a data science problem, it’s more of a business decision.” And you would not be wrong! Threshold selection lacks the appeal of say, generative adversarial networks.

Cryptocurrencies are useless. They’re only used by speculators looking for quick riches, people who don’t like government-backed currencies, and criminals who want a black-market way to exchange money. - Bruce Schneier

I would not give a fig for the simplicity this side of complexity, but I would give my life for the simplicity on the other side of complexity. - Oliver Wendell Holmes

The StreetEYE news aggregator experiment came to an end on 3/31/2019. Many thanks for supporting StreetEYE over the years!

Many people are talking about ‘quantitative tightening’ and ‘balance sheet reduction’, and some people are blaming it for market volatility, discussed here, here, here. IMHO, blaming balance sheet reduction for market volatility is cargo cult mumbo jumbo.

It’s been more than a year since we posted our last list of people to follow on Twitter for financial news. Time for an update!