Five Experiments to Beat the Sideways Market — and What the Data Actually Said

Our system beats the KOSPI handily over a 12-year backtest. But there is one regime where it barely makes money — the sideways market, which accounts for 60% of all months. We tried five different ways to fix this weakness, and all five failed. This is a record of those failures. It is also a story about why that failure turned out to be good news.

The Problem: 81 Sideways Months, +0.17% Per Month

Our regime-adaptive system (REGIME_C) splits the market into three states: bull, sideways, and bear. Each regime triggers a different strategy automatically.

Across a 12-year (136-month) backtest, here is the average monthly return by regime:

Regime	Months	Our System	KOSPI
Bull	21	+6.68%	+8.53%
Sideways	81	+0.17%	+1.49%
Bear	34	+1.65%	−4.32%

In bull markets we capture strong gains, and in bear markets we defend successfully even as the index collapses. But in the sideways regime — 60% of all months — we earn just 0.17% per month. Annualized, that is roughly 2%, effectively worse than a savings account. And we carry full market volatility and drawdown risk through all 81 of those months.

This looked like an obvious weakness. So we set out to fix it.

Experiment 1: Just Buy the Index in Sideways Markets

The simplest idea first. If the KOSPI returns 1.49% per month in sideways markets while we earn 0.17%, why not abandon stock selection and simply track the index?

The result was the opposite of what we expected.

Metric	Current	Index Tracking
Sideways monthly return	0.17%	0.06% (worse)
Full Sharpe	0.59	0.49 (worse)
Full max drawdown	−26.8%	−37.1% (worse)

Tracking the index made things worse. The culprit was a statistical trap. The 81 months we classify as "sideways" and the periods the KOSPI itself considers sideways are not the same windows. Our system demotes to "sideways" those stretches where the index rises but only a handful of large caps carry it — and during exactly those stretches, the KOSPI's average return was not 1.49% but 0.06%.

It also produced the classic illusion where short-term numbers improve while the full-period numbers deteriorate. The last three years happened to be a roaring bull market, so injecting the index into our sideways months simply borrowed that rally and looked good. Across the full 12 years, drawdown worsened by 10 percentage points.

Rejected. The defensive benefit of stock selection was wiped out by index beta.

Experiment 2: Buy Only the Relatively Cheap Names Within a Sector (Pairs Trading)

Experiment 1 taught us something. The returns in those 81 months were not in market direction (beta). So perhaps they were in the relative strength between stocks?

The data offered a clue. The stocks our system picks in sideways markets were almost all financials (banks and brokers), and the cross-sectional dispersion between them was twelve times the directional alpha. The market goes nowhere, but the individual names scatter widely. So what if we picked only the relatively undervalued name within a sector? This is the logic of pairs trading.

We tested it first with nine bank holding companies.

Method	Monthly Return	Cumulative
Hold all (current)	−0.61%	−31.7%
Cheapest only	−0.42%	−25.0%

"Cheapest only" did beat "hold all." But both lost money. It merely lost less — still a losing strategy. And the "relatively cheap" name scattered almost evenly across all nine stocks. If mean reversion were genuinely strong, the selection should converge on specific names; a different winner every month means the signal is closer to noise.

We swapped in brokerage stocks and tried again. It was worse.

Method	Monthly Return	Cumulative
Hold all	−0.17%	−16.7%
Cheapest only	−0.76%	−43.9%

The pairs filter actually amplified the losses. The names it flagged as "cheap" were mostly structurally weak small-cap brokers. It was not mean reversion — it was just repeatedly picking up weak stocks.

Rejected. Korean financials show weak, unstable mean-reversion relationships even within the same sector. And textbook pairs trading requires short selling, which — given the constraints retail investors face shorting in Korea — makes the strategy unsuitable for real-money operation in the first place.

Experiment 3: Avoid Only the Dangerous Sideways Months (Volatility Targeting)

If we could not find alpha, we would change tack and reduce risk instead. If we cut exposure only during the high-volatility "dangerous" stretches of the sideways regime, perhaps we could dodge losses while keeping the returns from the calm stretches.

For this to work, one premise had to be true: sideways losses had to be concentrated in high-volatility periods. We checked.

Volatility Tier	Avg Monthly Return	Win Rate
Low	+0.03%	52%
Mid	+0.48%	52%
High	+1.48%	56%

The result was startling. It was the exact opposite. In sideways markets, higher volatility meant better returns. The dangerous stretches were the ones making money. The dead zone was the low-volatility regime.

This finding solved an old mystery. We had once tried halving sideways exposure, then reversed it because it "only cut returns without defending against drawdowns." Now we knew why. The sideways returns were coming from the high-volatility periods, so cutting exposure meant amputating our best-earning stretch.

Rejected. Volatility targeting backfires by cutting off the source of returns.

Experiment 4: Empty Out Only the Dead Low-Volatility Stretch

Experiment 3 narrowed the target. The problem was not all 81 sideways months, but the low-volatility stretch where returns hover near zero. What if we emptied only that stretch into a risk-free asset (a money market fund), converting dead zero-return months into safe interest income?

This time we applied a strict rule that judges "is this low-volatility right now?" using only past data, never peeking at the future. With that constraint, the genuinely avoidable low-volatility stretch shrank to just 11 months.

Period	Current Sharpe	Low-Vol Avoidance Sharpe	Change
Full 12yr	0.697	0.684	−0.013
Mid	0.705	0.707	+0.002
Last 3yr	1.318	1.379	+0.061

On a full-period basis it actually got marginally worse. With only 11 avoidable months (8% of the total), the effect was negligible, and the slight improvement in the last three years was the same illusion we saw in Experiment 1.

Rejected. Even after precisely narrowing the target, the stretch was too small to produce any meaningful improvement.

Experiment 5: Just Move the Money Abroad (S&P 500)

By this point everyone thinks the same thing. "If the Korean market has no answer in that regime, just move the money to the US." It is the most natural objection. The idea is to park capital in the S&P 500 only during the low-volatility sideways stretch.

For this to work, a premise had to hold: when Korea is quiet, the US should be doing well. If Korea's low-volatility periods are when foreign capital flees to the US, then the US should be thriving in the meantime. Intuitively plausible.

We checked. Across the 11 low-volatility sideways months, we compared staying in Korea versus switching to the S&P 500. We included the exchange rate, calculating it as an unhedged US ETF that a Korean investor could actually buy.

Period (11 low-vol months)	Avg Monthly Return
Stay in Korea	+0.71%
S&P 500 (USD)	−0.04%
KRW/USD move	−0.50% (won strength)
S&P 500 (KRW terms)	−0.55%

Once again, the opposite. When Korea was quiet, the US was quiet too. Global markets move together far more strongly than we assume. Worse, the exchange rate hurt rather than helped: the won was actually strong in those periods, so holding dollar assets would have added a currency loss. The S&P 500 in won terms returned −0.55% — a full 1.25 percentage points per month worse than staying in Korea.

Rejected. Korea's quiet periods are not when foreigners depart for the US, but when the entire global market rests together.

There is a deeper lesson buried here. Our very premise — that the low-volatility sideways regime is "dead" — was actually false for the window we can identify in real time (11 months), where it returned a healthy +0.71%. It is easy to say in hindsight, "that was the dead zone," but in the moment you cannot know. There simply was no real-time-identifiable "dead stretch to empty out" in the sideways regime.

Conclusion: The Best Move Was to Touch Nothing

All five experiments pointed to the same place.

Experiment	Approach	Result
1	Index tracking	Worse drawdown, illusion
2	Pairs trading	Losses, weak mean reversion
3	Volatility targeting	Cut off the source of returns
4	Low-vol avoidance	Negligible effect
5	Switching abroad (S&P)	Pointless due to global correlation

The low returns across the 81 sideways months were not a bug to be fixed. They were a structural fact — that the Korean market grants little alpha and little beta in that regime. Every clumsy intervention reduced risk-adjusted returns without exception.

And stepping back, this was never a fatal weakness to begin with. Our system's 12-year risk-adjusted return (Sharpe 0.70) is nearly double the KOSPI's (0.37). The alpha comes from offense in bull markets and defense in bear markets. The sideways regime only needs to play the role of "hold the line without losing, even if you can't gain." The greed to win in every regime nearly ruined the system.

The best decision was to do nothing.

The hardest thing in quant investing is not building a good strategy — it is refraining from tampering with a good strategy you already have. We learned that after five rounds of verification. And we record the process without hiding it. This is how we work.

Bonus: During Those Quiet Months, What Actually Soared?

One curiosity remains. If our system was dormant in sideways markets, what rose the most in the broader market during those months? Could there have been a real source of returns we missed?

We checked. During the low-volatility sideways stretch where the system struggled most, we pulled the biggest one-month gainers across the entire KOSPI and KOSDAQ. The result was clear. 72% of the top names were KOSDAQ stocks, and the average one-month gain was 73%. The names tell the story at a glance.

Stock	Period	One-Month Gain	Theme at the Time
Anam Electronics	2017-02	+258%	Samsung-Harman deal, AI speakers
SY	2016-08	+238%	Solar / renewables
Isu Specialty Chemical	2023-05	+201%	EV batteries
Keum Yang	2023-06	+200%	EV batteries
Nature Cell	2018-08	+165%	Stem cell therapy

All were small- and mid-caps that rode a powerful theme (EV batteries, biotech, AI, renewables) to a 100–250% gain in a single month. It looks dazzling. But what came next matters.

Keum Yang soared more than 30-fold in a single year from around 5,000 won in July 2022, reaching a market cap of 11 trillion won. The fuel was not earnings but story. A mining venture that was supposed to generate 400 billion won in annual revenue was later revised down to tens of billions, and amid a false-disclosure controversy the stock crashed below 10,000 won. Some 240,000 minority shareholders were trapped. Nature Cell rose nine-fold in five months on hopes of regulatory approval for a stem cell therapy, then plunged when the approval was rejected.

Here is the final piece of our story. Even in those quiet periods, the market did hold enormous gains. But they lived in the realm of thematic speculation — unverifiable, unrepeatable — and most ended in a crash following the surge.

It is not that we failed to make money in sideways markets. It is that the returns of those periods lived in a realm we have decided, by discipline, never to touch. You cannot know in advance which stock will explode, and after it does, most collapse. We operate only with verified large caps and repeatable signals. As long as we keep that principle, these stocks are not ours to take.

It was not an absence of alpha. It was alpha we chose not to take. And we believe that choice was right.

---

All figures in this article are from a 2014–2026 backtest, reflecting 0.35% round-trip transaction costs and survivorship-bias correction. The individual stock movements in the bonus section are facts based on disclosures and press reports, and are not a recommendation to buy or sell any security. Many of the surging stocks mentioned subsequently suffered steep declines. Backtest results do not guarantee future returns.