Your First Alpha

Your teacher writes this on the board and reads it aloud:

Discuss in pairs (7 min), then share with the class (5 min):

1. What do you know how to do — from your country, your family, your work history — that no one in Newark has recognized yet?
2. In Lesson 06, China had 24,454 participants and a score of 53 million. The USA had 2,331 participants and 674,000. What is the difference between those two communities of researchers? Is it intelligence? Resources? Time? Or is it something else?
3. If every student in this classroom submitted one passing alpha today, what would happen to the USA score? Do the calculation together.

Your teacher wrote this lesson with Claude. The Principia Orthogona papers were edited with Claude. The Lean 4 proofs were reviewed with Claude. Claude is not a search engine and not a calculator. It is a thinking partner. It makes mistakes — you need to verify what it says. But it will help you understand things faster than any textbook, in your own language, at your own pace.

Open claude.ai now. If you use Chrome, install the Claude Extension — this lets you use Claude on any webpage, including the WorldQuant simulator, without switching tabs.

You have seen these operators in every lesson this semester. Ahmes used them to compress a circle into a square. Galileo used them to fold Ptolemy’s universe into Copernicus’s. A cell uses them to decide whether to eat itself. Today they appear in a financial simulator.

Go to platform.worldquantbrain.com/simulate.

Step 1 — Set your universe (Operator C).
In the right panel: Region: USA  ·  Universe: TOP3000  ·  Language: Fast Expression  ·  Neutralization: Subindustry  ·  Decay: 4  ·  Delay: 1  ·  Truncation: 0.08  ·  Pasteurization: On

Step 2 — Write a formula (Operator K). Type one of these into the code editor. Start with Alpha 1:

Step 3 — Simulate (Operator F). Click Simulate. Wait 3–5 minutes. The backtester tests your formula against years of real data. It will stall near 35% for a while — this is normal. Do not click away.

Step 4 — Read your results. When simulation finishes, click Show test period to reveal the TEST stats. Then scroll down to find IS Testing Status.

Step 5 — Check IS Testing Status. Expand the green PASS and red FAIL sections. You need 8 PASS / 0 FAIL. After the first 7 checks, click Check Submission to trigger the self-correlation check. It will show PENDING for a moment.

Here is what a perfect result looks like — all 8 checks green:

Step 6 — Submit (Operator U). If all 8 checks pass, click Submit Alpha. Your alpha enters the WorldQuant live pool. The USA aggregate score goes up. You just contributed to a global scientific competition from a classroom in Newark, NJ.

This section is about something that happened in the real research that built this lesson. It is one of the most important things you can learn about quantitative research — and about problem-solving in general.

What is self-correlation? Imagine you already submitted a recipe for chocolate cake. Now you try to submit another recipe that is 99% the same — just slightly more vanilla. The competition says: “This is not a new idea. It is the same idea with a small change.” Self-correlation measures this. If your new formula produces rankings that are 70% or more similar to a formula you have already submitted, it fails.

Why does this matter? The original submitted formula used this price signal:

Every formula that tried to replace this with a similar z-score — using different time windows (3 days, 5 days, 63 days), different price data (open, vwap), different volatility measures — produced self-correlation between 0.87 and 0.9987. All failed. The machine recognized the family resemblance even when the numbers changed.

The breakthrough: switching from a parametric z-score to a non-parametric ranking operator.

The winning formula — submitted and now ACTIVE, contributing to the USA IQC score:

What each part does, in plain English:

• sales / close — a “value” signal. Companies with high annual sales but a low stock price might be undervalued by the market.
• -ts_rank(close, 3) — a “reversal” signal. If the stock price is ranked lowest in the last 3 days, it may be due to bounce back. The minus sign means: bet on the lowest-ranked stocks going up.
• group_neutralize(…, subindustry) — removes the effect of the whole industry going up or down. We only want signals that are specific to individual stocks.
• ts_decay_linear(…, 4) — gives more weight to recent information. Yesterday’s signal matters more than last week’s.
• rank(…) — converts everything to a ranking from 0 to 1 across all 3,000 stocks.

Record your results below. The first two rows show real submitted alphas from the research that built this lesson — with their actual metrics. Rows 3–5 are for you to simulate yourself. In the last row, invent your own formula. Use Claude to write it in the correct syntax.

Option A — Instructions: Write a step-by-step guide for a classmate who missed today. How do you open the simulator? How do you set the universe? How do you write a formula? How do you know if it passed all 8 IS checks? What does a self-correlation failure mean, and what do you do about it? How do you submit? Use the imperative: “Open the website. Click on…” Use at least 8 vocabulary words from Stage 2. This is a real document a real person will use.

Option B — The Idea Behind Your Alpha: Explain your own alpha (the one in the last row of the practice table). What did you observe about human behavior that your formula is trying to capture? Why do you believe it might predict stock movement? What did Claude say when you described it? Did the backtester agree or disagree with your intuition? If it failed, what did the machine tell you, and what would you try next?

Option C — The Question That Does Not Have a Clean Answer:
The WorldQuant Brain platform does not know your name. It does not know where you were born, what your English level is, whether you are documented, or whether you grew up in a place with access to good schools. It only evaluates your formula. You have spent this semester reading research papers, building Lean proofs, learning contact geometry and autophagy and Fibonacci sequences — all of it in a second language, in a classroom at Newark, NJ.

Write your honest answer to this: What do you have to offer that the platform is waiting to discover? This is not a grammar exercise. Write what you mean.

Option D — The 30 Attempts (for stronger students):
The formula rank(ts_decay_linear(group_neutralize(rank(sales/close) + rank(-ts_rank(close,3)), subindustry), 4)) was the result of more than 30 attempts, two sessions, and one key insight: that changing window sizes on a z-score formula is not a structural change. It is the same idea with different clothes.

Write about a time in your own life when you kept trying variations of the same idea and kept getting the same result. What was the structural change that finally worked — or that you still have not found yet? You can write about language learning, about work, about a relationship, about anything real.

1. Install the Claude Chrome Extension. Use it on the WorldQuant platform to explain any operator the moment you see it. Try highlighting group_neutralize and asking Claude: “What does this operator do, and why does neutralizing by subindustry matter?”
2. Start a Claude conversation: “I want to become better at writing WorldQuant alphas. Can you teach me one new operator per day, starting with the simplest, with a plain-English explanation and an example formula using only: sales, close, volume, returns, assets, income?”
3. Try building a formula that combines a fundamental signal (a company metric divided by price, like assets/close or income/close) with a non-parametric price signal (ts_rank rather than ts_delta/ts_std_dev). Wrap it in group_neutralize(…, subindustry) and ts_decay_linear(…, 4). This is the pattern that worked.
4. Check the IQC leaderboard once a week. Track the USA score. If you submitted an alpha, you are in that number.