Why Microseconds Are Money
Low-Latency HFT Network Engineer · Module 1: Latency From Zero
Lesson 3 of 7
Prerequisites: What "Latency" Actually Means, What Trading Is (and Why Being Fastest Wins)
What you'll be able to do: put a dollar value on a microsecond of speed, explain in one paragraph why arriving second costs money, and judge whether an expensive faster connection pays for itself.
Suppose you learned tomorrow's winning lottery numbers one second before anyone else. That single second would be worth millions — not because of anything you built, but because you knew first. Now shrink that idea: the "lottery" runs hundreds of times a day, and the prize is real money changing hands in a market. The companies that know each result a millionth of a second before their rivals get first pick of every prize, all day long. This lesson puts numbers on that advantage: what one millionth of a second is actually worth, and why companies spend millions of dollars to buy it.
Here's the puzzle.
Scenario. You're advising a trading firm that's considering renting a faster private connection for $2,000,000 per year. The sales pitch says their 2-microsecond speed advantage will win them a profitable trade setup that occurs 50 times per day. Your job: run the numbers and tell them whether the connection pays for itself. No gut feelings — arithmetic only.
Given artifacts. One worked exhibit from the firm's own records, describing a single typical event. Assume 252 trading days per year.
EXHIBIT — a typical winnable event (from the firm's logs)
---------------------------------------------------------
A stock's price moves by $0.01 (one cent) while the firm's
old quote is still standing. The quote covers 10,000 shares.
Value of winning this event once:
$0.01 × 10,000 shares = $100
Events like this per day: 50
Firm's win rate with the 2 μs edge: 60% (they arrive first
in 6 out of every 10 events)
Trading days per year: 252
Cost of the faster connection: $2,000,000 / yearYour task: Compute the dollar value of the 2-microsecond edge per day (show the math), then answer in one sentence with the numbers whether the $2M/year connection pays for itself.
Workspace. Analyze-and-answer: a text box for your daily-value math and your one-sentence verdict. Submitting is optional — the ritual below is what unlocks the worked answer.
Hint ladder.
Hint 1 — where to look
Work strictly forward from the exhibit: value per event → events per day → fraction actually won. Don't touch the $2,000,000 until the very last step — it's the comparison, not part of the daily value.Hint 2 — what to compare
Daily value = ($100 per event) × (50 events) × (60% won). Compute that first. Then scale to a year with 252 trading days. Only then set the yearly total next to $2,000,000 and ask which is bigger.Hint 3 — the mechanism
The 60% is the part everyone forgets: a speed edge doesn't win every event — rivals sometimes still beat you, or the price moves the wrong way. You only bank six-tenths of the theoretical maximum. Multiply it in, don't skip it.Commitment ritual. ☐ "I've attempted this challenge and thought it through." Check the box (or submit an answer above) and the worked answer in S7 reveals. Nothing is graded; the struggle is the point.
Checking the box reveals the worked answer in S7 below. Returning learners stay unlocked.
The stale price on the shelf
Picture a shop that reprices its goods every few seconds. Between repricings, the shelf price can go stale — the world moved on, but the tag didn't. Now imagine you're allowed to buy at the shelf tag no matter what. If you notice the new real price before the shop updates the tag, you buy cheap and pocket the difference. That's the entire game: a stale quote (in plain English: a published price that no longer matches reality, in the brief moment before it's updated) is free money to whoever acts on it first.
In electronic markets, prices go stale constantly — hundreds of times a day per stock — because new information arrives in waves and every participant updates at a slightly different moment. The firm that processes each wave a microsecond sooner buys at the old price and sells at the new one, over and over. Each win is tiny — a cent per share — but the machine never sleeps, never hesitates, and never misses a wave it can see.
Why this matters for the challenge: the exhibit's "$0.01 on 10,000 shares = $100" is one stale-quote win. Your daily total is just counting how many of those the firm actually collects.
Why second place pays instead of getting paid
Here's the darker half, and it deserves one careful paragraph. When a firm's price goes stale, it doesn't just miss a profit — it becomes the victim. Some other faster firm sees the new reality first and trades against the stale price, which means the slow firm sells too cheap or buys too dear. This is called adverse selection (in plain English: the market systematically picks off whoever is slowest — your resting prices get "selected" by faster players exactly when they're wrong). Arriving second isn't neutral; it's a bill. Every microsecond of delay both shrinks your wins and grows your losses, which is why firms don't treat speed as a luxury. It's defense as much as offense.
Why this matters for the challenge: the 60% win rate already bakes this in — the other 40% of events are ones where the firm is the slow one, paying instead of collecting. Never assume a speed edge wins 100%.
Small, many, always
Three numbers multiply into the business: how much each win is worth (small — cents per share), how often winnable events happen (many — dozens to hundreds per day per stock), and how often the edge actually wins (always a fraction — rivals are fast too). Firms think in expected value: value-per-event × events × win-rate, exactly the arithmetic in your challenge.
This multiplication is why otherwise sane people spend millions to save microseconds. A 2-microsecond edge sounds like nothing — until it's the difference between winning 60% of fifty $100 events a day versus winning 40%. That 20-point swing is $1,000 a day, $252,000 a year, from one stock, from one tiny setup. Real firms run this math across hundreds of stocks and dozens of setups. The microsecond is small; the multiplication is enormous.
Why this matters for the challenge: your daily-value formula is the industry's core equation. Get comfortable with it — you'll use its shape for the rest of the track.
The rent-versus-winnings test
Every speed purchase — a straighter fiber cable, a faster connection, a computer placed physically closer to the exchange — is really a bet: will the extra winnings exceed the rent? Firms run this test ruthlessly, because the rent is certain and the winnings are probabilistic. A $2M/year connection that generates $3M/year in edge is a triumph; the same connection generating $750K/year is a donation to the cable owner.
Notice what this implies: there is no universally "worth it" speed. The same 2-microsecond edge is worth a fortune on a busy, jumpy stock and almost nothing on a quiet one. Speed is priced per opportunity, not per microsecond. When someone in this industry says "latency is money," they mean this specific multiplication — not a slogan.
Why this matters for the challenge: the final step is the rent-versus-winnings test. Yearly edge on one side, $2,000,000 on the other. Let the numbers speak.
- Step 1 of 4: Both firms see $50.00 and send buy orders. The price is the same for everyone — for now.
- Step 2 of 4: While the orders travel, the market price drops to $49.99. Anyone still quoting $50.00 is now holding a stale quote.
- Step 3 of 4: The fast firm's order arrives before the drop registers — it buys at the old price and pockets the cent-per-share difference.
- Step 4 of 4: The slow firm's order arrives after the move. The $50.00 price is gone; worse, if the slow firm was the one quoting $50.00, it just got picked off. Second place pays.
🔒 Revealed after the commitment ritual in S2 — attempt the challenge first. (Honor system: the page hides this until you check the box.)
Step 1 — value per day. Follow the exhibit forward:
- One event won = $100.
- 50 events per day × $100 = $5,000 of theoretical daily value.
- The 2 μs edge wins 60% of them: $5,000 × 0.60 = $3,000 per day.
Step 2 — value per year. $3,000 × 252 trading days = $756,000 per year.
Step 3 — the rent-versus-winnings test. $756,000 of yearly edge < $2,000,000 of yearly rent. Verdict: no — the $2M/year connection does not pay for itself; it costs $2,000,000 and returns only about $756,000 a year, so the firm would lose roughly $1.24M a year on the deal.
Wrong turns, named. Skipping the 60% — the most common error: $100 × 50 × 252 = $1,260,000 still loses to $2M, but the mistake matters because it inflates the edge by two-thirds. Comparing $3,000/day against $2,000,000/year directly — mismatched time units; the rent is yearly, so the winnings must be yearly too. Assuming "faster always pays" — the whole point of the exercise: speed is priced per opportunity, and this particular edge, on this particular setup, is overpriced.
Verify it worked. Sanity-check the scale: $3,000/day × ~250 days ≈ $750,000/year — yes, comfortably under $2M. And flip it: the connection would need to generate $2,000,000 ÷ 252 ≈ $7,937/day, i.e. a win rate of about 159% — impossible. The math confirms the verdict twice.
Check yourself — nothing here is graded. Wrong answers are the useful ones; each explains why.
Question 1. In one paragraph, what is 'adverse selection' for a slow trading firm?
Question 2. A different setup: each win is $40, events happen 200 times a day, and the firm's edge wins 50% of them. What's the daily value — and is a $1M/year connection worth it at 252 trading days?
Question 3. Why would a sane company spend $2 million a year to save 2 microseconds?
- A stale quote — a price that hasn't caught up with reality — is free money to whoever acts on it first, and markets create hundreds of them daily.
- Adverse selection means arriving second isn't neutral: your stale prices get picked off by faster firms, so slowness both shrinks wins and grows losses.
- A speed edge is worth (value per event) × (events per day) × (win rate) — never skip the win rate, since no edge wins 100%.
- Every speed purchase passes the rent-versus-winnings test: certain yearly cost against probabilistic yearly edge.
- Speed is priced per opportunity, not per microsecond — the same 2 μs is a fortune on a busy stock and pocket change on a quiet one.
Next: The Internet's Plumbing: Where Packets Actually Go — you can now price a microsecond; next you'll trace the actual journey a message takes, hop by hop, to see where those microseconds hide.