Korea capped 2x leverage. Retail bought 3x instead.
Korea tightened rules on 2x single-stock products and $1B more flowed into US 3x ETFs. It is a systems design failure, and engineers ship the same bug.
Korean retail investors put $4.63 billion into a single 3x leveraged ETF in the three weeks after their government tightened the rules on 2x domestic products. That is more than seven times what went into the next most-bought US stock. The regulator closed one door and the money walked straight to a riskier one.
I want to write about why, because the interesting failure here is not financial. It is a systems design failure, and I have shipped the same bug in software more than once.
🔍 What the numbers actually show
The reporting is from The Asia Business Daily (asiae.co.kr, 5 August 2026, by Park SeungUk), citing Korea Securities Depository data. I am commenting on it, not reproducing it, so the figures below are the ones that carry the argument.
Between 16 July (when the Korean government announced its single-stock leverage measures) and 3 August, the most-purchased US-listed security by Korean investors was SOXL, Direxion's 3x daily semiconductor ETF.
| Security | Tracks | Daily leverage | Purchase volume (16 Jul – 3 Aug) |
|---|---|---|---|
| SOXL | US semiconductor index | 3x | $4,636.17 M |
| Micron Technology | Itself (ordinary share) | 1x | $515.35 M |
| SOXL + KORU + TSLL combined | Chips, KOSPI, Tesla | 3x / 3x / 2x | $5,174.40 M |
| Same three, previous 13 trading days | — | — | $4,145.87 M |
Three of the ten most-bought US securities were leveraged ETFs. Combined buying rose by more than $1 billion against the prior 13-day window. KORU is a 3x product on the KOSPI itself, which means Korean investors were routing through New York to take triple-leveraged exposure to their own domestic index.
📉 The control worked perfectly. The goal did not.
The measures themselves landed exactly as designed. Minimum margin for domestic single-stock leverage went from ₩10 million to ₩30 million, and a 20% per-person portfolio cap took effect on 29 July. Domestic trading in the 16 affected products then did this:
| Date | Individual investors | Foreign investors |
|---|---|---|
| 30 July | ₩5,543.1 bn | ₩4,858.8 bn |
| 31 July (₩30m margin live) | ₩423.7 bn | ₩1,016.4 bn |
| 4 August | ₩309.2 bn | ₩488.0 bn |
Individual trading fell roughly 94% in three sessions. If your dashboard metric was "domestic single-stock leverage volume," you shipped a triumph.
Meanwhile, on 31 July, the first day the higher margin applied, SOXL was still the top US buy at $498.42 million, KORU second at $63.8 million, and those two alone made up 80.2% of the value across the top ten purchases.
The metric measured the mechanism, not the outcome. Volume moved. Risk exposure did not fall, it increased, because the offshore substitute is leveraged 3x rather than 2x.
🛠️ Every engineer has shipped this bug
Strip the finance out and this is a familiar shape: you constrain one path, demand is unchanged, and traffic finds the cheapest remaining route. The route it finds is usually the one you instrumented least.
| The control | What it assumed | Where demand actually went |
|---|---|---|
| Higher margin on domestic 2x | Cost suppresses appetite | Offshore 3x, no margin rule |
Rate limit on /api/v1 |
That is the only entry point | /api/v2, still open |
| Free-tier quota per account | Accounts are scarce | Ten free accounts |
| Client-side validation | Users use your client | curl |
| Blocking a domain at DNS | Name resolution is the chokepoint | The raw IP |
Each time, the limit binds where you can observe, appetite is unchanged, and the substitute path is worse precisely because nobody was watching it.
Key takeaway: A constraint does not reduce demand. It reprices one route. If you have not measured the substitute route, you have not reduced risk, you have only moved it somewhere you cannot see.
An industry figure quoted in the same report made the second-order point, that regulation arriving within two months of a product's launch creates its own risk:
"It's important to establish appropriate measures, but hasty responses may end up causing unintended capital outflows."
That is the credibility cost. In product terms: users who route around your limit rarely route back after you fix it.
🧮 Why 3x is not "one more than 2x"
Daily-reset leveraged products compound each day's move, so the multiplier applies to the path, not the destination. Take an index that drops 10%, then recovers fully:
// Index: 100 -> 90 -> 100. Net move: zero.
let x1 = 100 * 0.90 * 1.1111; // 100.00 — unleveraged, back to flat
let x2 = 100 * 0.80 * 1.2222; // 97.78 — 2x daily reset, down 2.2%
let x3 = 100 * 0.70 * 1.3333; // 93.33 — 3x daily reset, down 6.7%
The index is flat. The 3x holder is down 6.7% with no directional call gone wrong. Widen the swing and it gets uglier: a 50% drawdown needs a 100% gain just to break even, so a 3x product hit by a 20% down day is down 60% and needs +150% to get back to where it started.
- Volatility decay is not a fee you can shop around, it is the product working as specified.
- The cost scales with the square of daily volatility, so leverage is most expensive exactly when markets are chaotic.
- Nothing about "the chip sector goes up long-term" survives contact with this if the path is rough.
If you want to see what different return paths do to an ending balance, our investment return calculator and compound interest calculator make the compounding visible with your own numbers.
🇱🇰 The Sri Lankan version of this story
A Sri Lankan reader cannot run the Korean trade. Under exchange-control rules, residents do not freely move money offshore to buy US-listed ETFs, and I will not pretend to quote current thresholds at you. What is not closed is the pattern, and we have lived it:
- Import restrictions and grey channels. Restrict a category at the border and demand does not evaporate, it reappears through informal routes at worse prices and zero consumer protection.
- Forex controls and parallel rates. Cap the official channel and price discovery relocates rather than stops.
- Your own product's quotas. If you run a free tier for Sri Lankan users on a small budget, the per-account limit you just added is a per-signup limit. Signups are free.
The practical version: instrument the substitute before you tighten the primary. Log what people do after they hit the wall. That log is the only evidence your control did anything real.
💡 What this means for you
- If you write policy, code, or pricing: define success as the outcome you care about, never as the throughput of the thing you throttled. Korea's domestic volume metric looks like a 94% win and describes a worse risk position.
- If you are tempted by leverage: 3x is not incrementally riskier than 2x, it is a different instrument whose losses compound against you on flat markets. Run the numbers on a path, not an endpoint.
- If you run a free tier: assume every limit you set will be met by a workaround within a week, and decide now whether you can live with that workaround.
- If you are a student watching this: the transferable skill is not stock-picking. It is asking "and then what happens?" one step further than the people who designed the rule did.
The Korean regulator was not wrong to act. Single-stock products that amplify daily moves genuinely do amplify market volatility. The gap is that the intervention was scoped to what the regulator could reach, while appetite was scoped to what the investor could reach. Those two boundaries were never the same shape.