Wealthfront Engineering

Keep in mind that Frequent A/B tests burn statistical “credit.” Any time you ship a winner at p = 0.05 you’ve spent 5 % of your false-positive budget. Do that five times in a quarter and the chance at least one is noise is 1 – 0.95⁵ ≈ 23 %.

There are several approaches you can take to reduce that source of error:

Quarterly alpha ledger

Decide how much total risk you want this quarter (say 10 %). Divide the remaining α by the number of experiments left and make that the threshold for the next launch. Forces the “is this button-color test worth 3 % of our credibility?” conversation. More info: “Sequential Testing in Practice: Why Peeking Is a Problem and How to Fix It” (https://medium.com/@aisagescribe/sequential-testing-in-practice-why-peeking-is-a-problem-and-how-to-fix-it-59d7024344f6).

Benjamini–Hochberg (BH) for metric sprawl

Once you watch a dozen KPIs, Bonferroni buries real lifts. BH ranks all the p-values at the end, then sets the cut so that, say, only 5 % of declared winners are false positives. You keep power, and you can run the same BH step on the primary metric from every experiment each quarter to catch lucky launches. More info: “Controlling False Discoveries: A Guide to BH Correction in Experimentation” (https://www.statsig.com/perspectives/controlling-false-discoveries-guide).

Bayesian shrinkage + 5 %

“ghost” control for big fleets FAANG-scale labs run hundreds of tests and care about 0.1 % lifts. They pool everything in a simple hierarchical…

Also former HFT / market maker here (UBS, GETCO), and also the developer who wrote Wealthfront's direct indexing with tax loss harvesting 10 years ago.

I had that same skepticism before I built it. Using a Bloomberg terminal back then, my conclusion was that the weighted spread for the S&P 500 was 3.2 bps, vs. 0.6 bps for SPY.And this was > 10 years ago, so I'd think by now it would be even tigher. The ratio may have changed, but who cares? It's like saying that rice got more expensive at the supermarket - it's already so cheap that it doesn't matter.

With tax loss harvesting specifically, each order typically has a threshold, so that you only trade when the projected tax benefit is a large multiple of the transaction cost.

Also, I'm sure this is obvious to you if you work in market making, but for others reading this: the spread costs aren't additive (re: 'every. single. stock'). If you have 500 stocks, each with 2 bps round-trip spread cost, but each is at e.g. 1 / 500 = 20 bps, then the weighted spread for the entire basket is 2 * 500 * 1 / 500 = 2 bps. It's not 2 * 500 = 1000 bps. The main question then is - how much tighter are spreads for ETFs than for the average stock? And, since bigger stocks (AAPL, NVDA etc.) will have tighter spreads than smaller index constituents, the weighted average will be even lower.

Here's my blog post:

https://eng.wealthfront.com/2014/03/04/marketside-chats-4-costs-of-replicating-an-sp-500-portfolio/

Wealthfront CEO Andy Rachleff on this:

"Ignore Company Size

Great engineering mentors reside at companies of all sizes. Coding best practices are broadly similar between both large and small companies, provided both are on the leading edge. That should come as no surprise because engineers need to solve many of the same root problems, no matter what the size of the company.

Failure at a small company with a great engineering team is not as risky as you might think. Even if your employer fails, recruiters are quite knowledgeable about which companies have recruited great technical leaders and are desperate to hire their disciples. Perhaps even more surprising is engineers from failed companies with great technical reputations are far more vigorously pursued than engineers from successful companies that lack killer technical reputations.

Once again, the most important consideration is the quality of people from whom you can learn."

http://eng.wealthfront.com/2014/02/career-planning-for-new-grad-and-young.html

Wealthfront | Redwood City, CA | Onsite | Visa

Wealthfront takes the guesswork out of sound, long-term investing through effortless automation. We efficiently build and deliver products which pave the way for a new generation of investors to achieve their financial goals. With their trust, we believe we can and will change this industry. Find out how our engineering team contributes to our mission at http://eng.wealthfront.com

We are hiring across the board, but are specifically looking for Sr. Backend Engineers with Java experience, Full Stack Engineers with JavaScript experience and Engineering Managers.

Feel free to check out the job descriptions and apply here: http://grnh.se/6regmv1 (please mention HN in application).

Wealthfront (formerly kaChing), Palo Alto CA (no remotes)

Wealthfront is intent on disrupting the $10-TRILLION mutual fund industry, an industry that's seen little innovation in the past 25 years. To reach this goal, we put our customers and our technology at the heart of everything we do. Java, Scala, JRuby on Rails. We have a 5-minute commit-to-production continuous deployment system. Read our eng blog at http://eng.wealthfront.com.

Lead UI Designer /
Frontend Software Engineer /
Backend Software Engineer

Email jobs@wealthfront.com /
More at: https://www.wealthfront.com/jobs

Wealthfront | Palo Alto, CA | Onsite | Visa | Full-time and Interns | http://wealthfront.com

We invest your money using software. We are a technology company, not a finance company. Read more about our engineering culture here: http://eng.wealthfront.com/

We are hiring for all sorts of positions (backend, iOs, and Android engineers to name a few) see them all here: https://www.wealthfront.com/careers

Email me directly: fabien (at) wealthfront (dot) com