Do Korean Stocks Rise When US Biotech Stocks Do? I Measured, It Was Different, and What It Costs (Building a Stock App, Part 3)
I assumed that when US biotech stocks rose, related Korean stocks would follow. But measuring about a year of prices, only one of five stocks rose clearly more often than usual. So I dropped the US-rise alert, and I handle the hourly checks with plain calculation instead of AI to keep costs down. I also share the day the numbers were wrong.

Do related Korean stocks rise when US biotech stocks jump? Measuring about a year of prices, only one of five stocks rose clearly more often than usual. The rest were about the same as usual, or had too few cases to judge. So I dropped the alert that would ring every time the US market rose, and made the app notify only when something actually changes in your holdings. The hourly checks also run on calculation without calling AI, so AI costs don’t grow with each check.

If US biotech stocks jump, won’t related Korean stocks follow? Even I, who had never invested in stocks, thought so. While building an app to help my older brother check his holdings easily, I thought alerting him to news like this would help.
But before building the alert, I decided to check one thing first. Did that really happen often?
I compared about a year of prices
I counted how often related Korean stocks rose on the next trading day after a US stock rose 3% or more in a day, and compared that with how often each usually rose. I’ll call the stocks A through E instead of using their names.
| Stock | Usually rose | Rose after a US rise |
|---|---|---|
| A | 45% | 69% (25 of 36) |
| B | 41% | 44% (27 of 62) |
| C | 48% | 47% (35 of 74) |
| D | 45% | 57% (8 of 14) |
| E | 43% | 50% (7 of 14) |
*Results compared on about a year of prices as of November 5, 2026.*
The biggest difference was A: 45% usually, but 69% after the US side jumped. B and C, by contrast, were almost the same as usual. D and E had only 14 cases each, too few to judge.
I expected companies closely tied by business to move together, but in the records I looked at, that wasn’t always so. And A’s difference doesn’t mean it will continue. It’s a past observation, not a test of predictive power.
So I dropped the US-rise alert
If I hadn’t measured, I would have built an app that rings your phone every time the US market rises. But there was no reason to send alerts for cases that barely differ from usual.
So US news is now only logged in the app each morning. Phone notifications go out only when something actually changes, such as a holding rebounding sharply from its recent low or rising 5% or more in a day.
These aren’t buy or sell signals either. They say, “take a look today.” The app is meant to let users worry less, so calling them back again and again over weak news would defeat its purpose.
What does running it every hour cost?
The app does its work at set times, even if no one opens it. On weekdays it runs in this order.
- 8:30 a.m.: logs the US market
- 9 a.m. to 3 p.m.: checks prices and alert conditions every hour
- 4:40 p.m.: runs a full daily check
- 5:10 to 5:40 p.m.: updates comparison candidates
Running this often sounds like it would rack up AI fees, but the hourly checks don’t call AI. They fetch prices, compare them with earlier values and calculate whether the set alert conditions are met. Comparing numbers is something an ordinary program can handle.
AI is used only for tasks that need it, such as recording predictions or reading trade-screen photos. Instead of handing everything to AI, I separated what calculation can do from what needs AI.
For now, the app server and database run within free plans. AI usage costs extra, but the setup doesn’t add AI costs every time prices are checked hourly. Of course, if usage grows, I’ll need to look at server and database costs again.
The cost-saving method I learned while building it was surprisingly simple: don’t use AI for work that doesn’t need AI.

Some days the numbers were wrong
Does handling things by calculation make everything accurate, then? Not quite.
There was a time when the prices shown in the app didn’t match the real ones. It was showing old saved values instead of fetching new ones. At first I was told they were at most five minutes old, but when I checked, that explanation was wrong too.
Ask AI for the reason and you get an answer quickly. But whether that answer was an explanation based on the real cause, or a plausible guess, had to be checked separately.
After that, I realized I need to look not just at a value but also at when it was checked. A number showing up neatly on screen was no guarantee that it was current.
AI greatly speeds up building. But it didn’t automatically finish the work of checking that things were built right.

For users to be lazy, the builder has to be diligent
For users to worry less, the builder had to compare things in advance and filter out unneeded alerts. Frequently checked tasks had to run automatically, and work that didn’t need AI had to be done by calculation.
That’s also where I spent much of my time building this app. Removing features without enough evidence mattered as much as adding new ones.
I think automating company work is similar. What to run automatically, what to tell people about, and what not to do at all can only be decided by looking at and measuring the real work.
What I want to build is not an app that makes users diligent, but one that helps them not miss the facts they need, even while worrying less.
In short
- After US biotech stocks rose, only one of five stocks rose clearly more often than usual the next day.
- So I dropped the US-rise alert and notify only when something actually changes in your holdings.
- Hourly checks run on calculation without AI, so AI costs don’t grow with each check.
- Even calculated numbers can be wrong, so I now look at when a value was checked, not just the value.
It’s the same when I build FROX, the business system at ONE-PEX. What to automate and what to leave out can only be decided by measuring the real work.
*This post is a record of how I built the app, not investment advice.*
Related reading: Can You Trust AI Stock Predictions? The Safeguards I Built So I Wouldn’t (Building a Stock App, Part 2) · A Stock App That Never Tells You to Buy or Sell (Building a Stock App, Part 1)

A software planner who became a developer after two years of working with AI five to twelve hours a day. Today I build FROX, an AI-powered business system, at ONE-PEX.
More posts
- What Can You Build with AI?Orders in KakaoTalk, Stock in Excel, Receivables in a Notebook: Changing Messenger Orders Only the Owner Understands (Food Supplier, Part 1)
- What Can You Build with AI?I Fixed My TWA App, So Why Didn't It Change? What I Learned from the War with Caching
- What Can You Build with AI?Details Matter: Splitting 3D Characters and Vehicles So They Can Move (Building Road Voca 2, Part 5)
