Two earnings trades planned with an AI assistant, and what happened next
In July 2026 a member asked Claude, connected to EarningsWatcher through MCP, for an earnings trade. Twice. One long straddle on NFLX, one short strangle on SOFI. He paper-traded both. We went back to our own data to see how each one ended. One won, one lost, and the loss is the more useful of the two.
The assistant did not guess. Every figure it quoted came from EarningsWatcher tools: the live implied move, the ten-year move distribution, the option chain, the backtester, the paper-trading book. What it could not do is decide, and the two results below show exactly where the deciding happens. Education only, paper money throughout.
1. NFLX, July 16: “what trade setup for earnings today?”
At 3:39 pm on the day Netflix reported after the close, the member asked the assistant which name to play and how. It scanned the day's reporters and came back with NFLX, for one reason: a gap between what options were pricing and what the stock usually does.
Implied move is 8.4%, but NFLX's 10-year average earnings move is 10.4%. The market is under-pricing the historical move, which is the classic long-volatility trigger. IV Rush is flagging it too.
Then it priced a structure off the live chain, expiring that Friday:
| Structure | Debit | Breakevens | Pays if |
|---|---|---|---|
| ATM straddle | about $6.59 | −10.2% / +9.3% | the stock moves more than about 9 to 10% either way |
| 77c / 71p strangle | $3.99 | −10.5% / +9.9% | a slightly bigger move, for less premium at risk |
| 80c / 68p strangle | $2.17 | −12.0% / +11.1% | a big move; cheap, but usually expires worthless |
The member asked the obvious follow-up: how much profit can I expect? The assistant ran the simulator with the IV crush applied and answered with a table instead of a promise. The line that matters:
Both structures are still roughly at breakeven if the stock only moves the implied 8.4%. You need the move to beat what is priced in. At the 10.4% average, the straddle nets about $40 to $96 per contract.
He pushed once more: backtest alternatives with better risk-to-reward. The honest answer was the best moment of the conversation. The backtester held only two NFLX events at the time, January and April 2026, and the assistant said so before showing anything: “two data points, not twenty.” In both, the long straddle lost 66% and 84% of its debit by the next close. The short strangle kept nearly all of its credit both times.
The tool the member asked for a long-volatility trade had just shown him that the same trade lost the last two times. That is what a connected assistant is for. It surfaces the inconvenient number instead of burying it.
2. What NFLX actually did
From our implied-move history for NFLX, the July 16, 2026 print:
| Measure | Value | Versus the trade |
|---|---|---|
| Implied move at the close | ±9.4% | It had risen from the 8.4% quoted at 3:39 pm. The gap the thesis relied on had half closed by the bell. |
| Peak intraday move next day | −12.5% | Beyond the straddle's −10.2% breakeven. A winner, if closed there. |
| Next-day close | −7.3% | Inside the breakeven. At the planned exit, the straddle lost roughly a fifth to a quarter of its debit, by the assistant's own P&L table. |
So the stock moved more than the historical average, and the trade still lost. The reason is not bad luck. The 10.4% average that built the thesis is a peak intraday figure, the way we compute the default earnings move across the site. The P&L table was computed at the next day's close. Those are two different statistics, and NFLX split them: it went through the breakeven during the session and came back inside it by the close.
Our wider record for NFLX says the same thing. Over the last 16 prints with an implied move on file, the close landed inside the implied range 8 times, but the intraday peak stayed inside it only 6 times. A long-volatility trader on Netflix is paid at the extremes of the session far more often than at the bell.
3. SOFI, July 29: “one idea with backtested results”
Two weeks later, a different question: what to trade next? One idea, with backtested results. The assistant went looking for the opposite edge, a name whose options habitually price more movement than the stock delivers.
Over the last 15 quarters, SOFI's implied earnings move has priced rich relative to what actually happened. The realized move stayed inside the strangle in 13 of 15 prints.
The backtest it showed used real historical fills: a short strangle at roughly 20 delta on each side, sold at the prior close, bought back at the open right after the report.
| Backtest, 15 SOFI reports (2022 to 2026) | |
|---|---|
| Win rate | 13 of 15 |
| Average return on premium | +32.6% |
| Worst two | −51.4% (July 2025) and −47.4% (November 2022), both jumps straight through a strike |
The member did not take the idea on faith. He said paper-trade this in EarningsWatcher now, and the assistant placed it in his paper book with the thesis written into the trade description:
| Leg | Strike | Delta |
|---|---|---|
| Short call | $20.00 | 0.22 |
| Short put | $15.50 | −0.20 |
- Net credit $0.65 per contract, breakevens $17.10 to $22.90, about ±17% from the stock.
- Implied volatility 77.5%, expected to crush to about 59% after the print.
- Plan: hold through the Wednesday pre-market report, close at the open, the same rule the backtest used.
4. What SOFI actually did
From our history for SOFI, the July 29, 2026 print:
| Measure | Value | Versus the trade |
|---|---|---|
| Open after the report | −7.4% | Well inside the ±17% range. This was the planned exit: the position was bought back with most of the credit kept. |
| Peak intraday move | −11.1% | Still inside, even for someone who ignored the plan and held through the session. |
| Close | −8.9% | Inside. |
A clean win, and a boring one, which is what a short-volatility trade is supposed to look like. Two things kept it boring. The implied move going in, about 12%, matched SOFI's average peak move of the last two years, and the closes have run well inside it. And the exit was immediate. The print before this one, in April 2026, closed −15.4% against a 10.2% implied move. The same structure, held through that session, would have been one of the backtest's losing rows.
Our record for SOFI over its last 16 prints with an implied move on file: the close stayed inside the implied range 12 times. The intraday peak stayed inside only 7 times. Sellers of SOFI premium are paid for being quick.
5. What the two results teach
Exit where the statistic lives
The NFLX thesis was built on peak moves and cashed at the close. That mismatch, not the direction call, decided the trade. Before entering a long-volatility position, decide whether the plan is to sell into the spike or hold to a fixed time, and check the historical number for that exit. Our implied-move pages show both the peak and the close for every print, for this reason.
Two data points are a warning, not a backtest
The assistant said it, and it was right to. NFLX has a forty-print move history on our pages and only two events in the option-level backtester, because the backtester needs full historical chains and those start in 2026 for most names. When the sample is that thin, the move history is the better guide, and the backtest is a sanity check on structure, not on edge.
Rich implied plus a fast exit is the short-vol recipe
SOFI worked because both conditions held. The 13-of-15 record contains two losses of about half the credit, and the print just before this one would have been a third. A high win rate on an earnings strangle describes the exit discipline as much as the stock.
The assistant computed, the trader decided
Neither conversation was “tell me what to buy”. It was: which name, what structure, priced how, and what does the history say. Then the member chose, papered it, and could review the result later. That is the workflow we built the MCP for, and it is why the losing trade on this page is not an embarrassment. It is the dataset doing its job.
6. How to run the same conversation
Everything above was done in chat, with paper money. To repeat it:
- Connect your assistant: Settings → AI Connectors (MCP) in the member dashboard, then paste the MCP URL into ChatGPT, Claude or Cursor and sign in. The MCP page has the steps.
- Ask for today's setup: “Which name reporting today has the biggest gap between implied and historical move, and what structure would capture it?”
- Make it show its P&L: “Simulate that at ±5%, at the implied move and at the historical average, IV crush applied.”
- Make it show the record: “Backtest the same structure on this name's past prints, exit at the next close, and tell me the sample size.”
- Before you take it: “Is the historical number a peak or a close? Match my exit to it.”
- Paper it: “Paper-trade this in EarningsWatcher with the thesis in the description.” It shows up in your positions and your journal.
Two more member stories: how one member turned the MCP into a personal trading OS, and the free Rewind game, where you can trade real past earnings weeks with the same data before risking anything.
Ask your own assistant the same questions
Membership includes the MCP: implied moves, move history, simulator, backtester, paper book and journal, in ChatGPT, Claude or Cursor.
