Economist Hong Hao said in a recent conversation that this year’s AI trade cooldown makes the most sense when you look at the brutal reversal in South Korea’s semiconductor stocks. To him, it is one of the cleanest case studies of how bubbles inflate and how they snap.
He organized the discussion around three questions: why the AI sector suddenly lost heat in the middle of the year, why South Korea’s semiconductor rally turned so violently, and why the market that seems to be making money for nearly everybody may be the one investors should fear the most.
South Korea’s semiconductor bubble cracked in about 40 days
Hong said one of the biggest things to happen in capital markets lately was the surge, peak, and break in the global semiconductor sector, with South Korea right at the center. From the top, the South Korean semiconductor trade was cut roughly in half in about 40 days, and the swings have stayed fierce ever since.
By his account, investors still sitting in South Korean semiconductor names have been under real strain, especially anyone using leverage. For leveraged players, he said, the market has felt like a daily whip between forced selling and sharp rebounds.
What made this look like earlier bubbles, he said, was messy leverage. People loaded up on borrowed exposure and drove the bubble line sharply higher. And that same process, in his view, all but ensured an equally fast drop. This time the climb and the collapse both happened without the longer buildup and unwind seen in some past cycles.
He said the relevant South Korean market gauge peaked near 9,000, then dropped hard toward 5,000 before bouncing. For anyone trying to understand cycles and bubble behavior, he said, that sequence is worth close study.
Hong said the core assumptions of traditional economics fail in real markets
Hong then went after the basic models taught in business schools, including the efficient market hypothesis, random walk theory, and expectation-based models. In his view, they all depend on assumptions that fall apart in actual trading.
The first assumption is complete information transparency, where every investor can easily get the same information. Hong said that is plainly false. Equity markets are full of negative information that does not get reflected in prices right away, and prices often move before news is officially released, which suggests some participants get information before the public does.
The second assumption is that all participants are fully rational. In that framework, investors are extremely sensitive to marginal cost and marginal return, they exit automatically when return no longer covers cost, and they can keep calculating the optimal allocation at all times. Hong said that is even farther from reality. Behavioral economics, he argued, shows that most people crack when overloaded with information and fall back on shortcuts. And at market extremes, those shortcuts are often dead wrong.
The third assumption is frictionless trading. Hong said trading costs are large, and the bigger the fund, the more each trade moves the market. He added that some quantitative funds often buy baskets of small-cap and micro-cap stocks while shorting an index to capture what gets called alpha, even though that alpha may just come from liquidity conditions.
His verdict was blunt: the pricing logic implied by efficient market theory does not work, and the real world may actually look like the reverse of those assumptions.
Gold rising with equities was, in his view, a rare regime-change signal
Hong said the market flashed a striking pattern in November last year: gold and equities went up at the same time, something he described as not having happened in the previous 60 years.
Gold, he said, was rising in a parabolic move while stocks were also surging across the US, emerging markets, and Europe. That pairing stood out because gold is usually treated as a hedge asset in a portfolio, the thing that offsets equity stress because its price path is not supposed to move in lockstep with risk assets.
And it was not just gold. Other precious metals climbed alongside equities too. According to Hong, that sort of setup was last seen in the late 1970s and early 1980s, around the collapse of the Bretton Woods system and the creation of a dollar-centered fiat credit regime.
When gold and stocks rise together, he said, the market is moving from one regime to another. In that kind of environment, investors may find that the tools they thought would hedge their portfolios no longer do the job.
Traditional portfolio construction assumes that mixing assets with low correlation can smooth volatility. Hong said that breaks near turning points and during regime changes, when correlations shift. In concentrated risk events, correlations across assets tend to converge, and portfolios lose the hedge effect they were built to have.
He described the previous four decades as an era shaped by China’s entry into the World Trade Organization and a jump in labor productivity, forces that pushed global inflation structurally lower. In that world, every major selloff looked like a buying chance because falling inflation gave the Federal Reserve and other central banks room to keep cutting rates. He pointed to interest rates falling from the teens in the 1980s to around 1% in the 2000s, then to 0 during the 2008 quantitative easing period, with Europe even going negative. In his view, that long regime ended after 2020.
The 2022 inflation shock broke the old 60/40 playbook
Hong said quantitative easing intensified after 2020. It was not just central-bank liquidity expansion; the US Treasury was also sending money straight to households, flooding the system with supply. Against that backdrop, US inflation hit 9.1% in 2022, a level he said the market had not seen in more than 40 years.
He also made a point about industry experience. Finance is a hard business, he said, and not many analysts stay in the field past age 40. So a large share of market participants had never actually lived through an inflation shock above 8%. When the regime changed, many portfolio managers were still leaning on the stock-bond 60/40 framework that had worked over the previous 14 years, only to watch it fail badly in 2022.
For Hong, that was a plain example of the danger of applying an unchanged reading of modern history to a world that is already shifting.
Markets reward trend followers and punish those who fight the tape
Hong split market participants into several broad trading types. First came the momentum trader. Simple version: the strong get stronger. He added that price momentum can absorb only so much capital, yet most quantitative funds he knows rely on trend-following strategies. Once everybody uses the same factor in allocation, the trend starts feeding itself.
He called these traders convergence traders and said they are the market mainstream. Before a trend hits an extreme, he said, momentum strategies usually work. He added that after the quant collapse in 2024, these strategies were still effective through the first quarter of 2025 and 2026, then broadly broke down in the second quarter.
The second group was the contrarian, or divergence trader, the investor who believes in mean reversion. Hong said a cycle from low to high and back to low usually takes three to four years, which is why quantitative value strategies suffer a collapse every two to three years. Once the window opens and the trend reaches an extreme, mean reversion follows.
Even so, he said, market structure favors momentum trading, not real criticism of consensus. Fighting the market is exhausting. Before a bubble bursts, he said, contrarian investors are often insulted and humiliated because they are seen as getting in the way of easy money.
The third group was the random trader. No model. No fixed framework. No consistent process. They buy because prices are going up and sell because prices are going down. Hong said their presence magnifies disorderly moves, and when they act in parallel with convergence traders, bubbles become easier to build. He cited that mix as one reason the semiconductor bubble inflated so fast.
Why retail investors often end up buying the top
Hong gave a simple example. A group of retail traders may have no idea which sector is attractive and just buy whatever everyone else is buying and sell whatever everyone else is selling. He said that helps explain why retail money so often shows up at the top.
He argued that even if some investors correctly spotted the South Korean semiconductor bubble, that does not mean most participants got in early enough to make money from it. A market can run from 4,000 to 9,000 and then drop back to 5,000, leaving some traders still ahead on paper, but very few actually bought near 4,000. Random traders, in his view, almost always enter near the peak.
That is why turnover tends to bunch up near tops while buying at the bottom stays limited. To Hong, that is one more reason the fully rational trader imagined by efficient market theory does not exist in real life.
Low volatility over time creates fertile ground for leverage
Hong said investors often talk about wanting a slow bull market. But if a market is really rising around 8% a year with only small drawdowns while the 10-year rate sits near 1.8%, then borrowing at 1.8% to buy an asset compounding at 8% becomes an obvious trade.
Under those conditions, he said, leverage rises. In a market where volatility stays suppressed for a long time, leverage does not remain flat.
He pointed to several sources of leverage in South Korea: hedge funds on both the long and short sides, retail margin financing and derivatives, and the issuance of leveraged ETFs. One example he gave was 7709, a Hong Kong-listed product offering 2x long exposure to SK Hynix. Hong said it rose 12-fold in one year, but it was also one of the main vehicles through which investors got wiped out and pushed into liquidation.
He also said South Korea approved leveraged ETFs in June, when the market was already trading at elevated levels. In his telling, part of the logic was that Hong Kong’s 2x long ETF had once been the world’s largest single-stock leveraged ETF, and South Korea did not want that money being made somewhere else. So the local market introduced 2x and 3x long products tied to Hynix and Samsung.
For Hong, opening the door to leverage tools at historical highs is a major trading error. In a market that looks calm while price expectations keep rising, leverage grows. Then rising prices give momentum traders and random traders the easiest possible reason to keep buying: the market is already going up.
He also explained why the concentration became hard to sustain. Samsung and SK Hynix make up about 20% of the South Korean market, he said. If those names triple in a year, their combined weight can move above 50%. Push that further and there is barely any room left for the rest of the index. That, he said, cannot go on forever. He compared the current path of South Korea’s KOSPI with past tech-stock bubbles and said the market collapsed once Samsung and SK Hynix rose to more than 50% and close to 60% of the index.
Three conditions are needed for a bubble to form
Hong said bubbles need three things.
- First, a grand narrative. The bigger the perceived technological revolution, the stronger the story about future productivity and income growth.
- Second, leverage or regulatory loosening.
- Third, investor profit-seeking behavior, which speeds up price chasing when expectations are high and financing is easy to get.
Put those together with abundant money and a market that feels calm on the surface, and bubbles are much easier to build, he said.
He used insurance pricing as an analogy. If ships moving through the Strait of Hormuz pass through a long quiet stretch, insurance may stay stable at a few hundred thousand per ship because pricing models are based on what has looked safe for years. But in the insurance business, he said, the quietest stretches and the lowest premiums can be exactly when vigilance matters most.
Hong said markets work the same way. When different kinds of traders all start expecting the same thing, leverage and bubbles build together. By June, he said, almost no investors willing to take a counter-cyclical view or speak against the crowd were left. And that, in his view, is when a bubble becomes most dangerous.
The illusion of diversification
Hong also challenged a common idea in portfolio management: that holding a big basket of stocks automatically means the risk is diversified. He said if the underlying source of risk is the same, then the portfolio may not be diversified at all.
He described seeing investors who hold 30 or 40 stocks, sometimes more, yet cannot clearly explain what many of those companies actually do. Whether it is one name or another, or even a company like MiniMax, the answer often comes back to the same line: it is a leading AI company. But if all those stocks share exposure to the same semiconductor or AI cycle, then they also share the same risk factor.
In that case, adding more positions does not reduce portfolio volatility. It may even raise it, because the losses can hit from the same source at the same time.
He compared that setup to a forest. When trees are sparse and still growing, wildfire risk is limited. Once the forest gets dense and the branches tangle together, one lightning strike can destroy the whole thing. Periods that look calm and prosperous, he said, can also be the periods when risk is building fastest.
What comes after a bubble bursts
Hong said the speed of the South Korean bubble’s collapse was, based on available data, the fastest on record, with half the value erased in 40 days. In a product such as the 2x long Hynix vehicle 7709, he said, the price could drop from 200 to 20, a 90% fall, and then rebound 70% in a single day.
After the first break, he said, bubbles usually move into a consolidation and digestion phase. By his estimate, the adjustment range is often about 50% to 65%.
He compared that with larger equity markets such as the Nasdaq after March 2000, when the drawdown reached roughly 80% at its deepest point. For a buy-and-hold investor, he said, it took more than 20 years to get back to the March 2000 high.
That is why, in his view, it is not enough to say a stock will make money no matter what. Sequence matters. Entry point matters.
A bubble can burst without ending AI investment
Even after spending much of the conversation on leverage, crowding, and collapse, Hong did not paint bubbles as purely destructive. He said bubbles are one of the strongest tools for changing human society. Without a large bubble, it would be hard to mobilize enough labor, capital, and infrastructure investment to bet on the future, including AI hardware and cloud-computing centers.
He said the outline of that future world is already visible. More GPUs will be needed for inference, more storage will be required, and more materials will be needed for robotics. A collapse in AI-related stock prices, he said, does not stop that process.
Instead, Hong said, new models, new investment, and new opportunities can emerge from the wreckage after the bubble breaks.
This article is from the WeChat account "Capital Deep Dive," authored by Capital Deep Dive.