Most quantitative research focuses on US equities, European bonds, or a handful of developed-market indexes. That makes sense given data availability, liquidity, and institutional infrastructure. But it leaves a question unanswered: do the basic premises of systematic trading hold in markets that look nothing like the S&P 500?

Poudel and Paudel (2025) attempt to answer this for what may be the smallest stock exchange featured in recent quant research. The Nepal Stock Exchange, NEPSE, has a single index, no short selling, limited institutional participation, and a trading culture dominated by retail investors acting on tips, social media hype, and what the authors bluntly describe as Fear of Missing Out. Their study builds a rule-based quantitative strategy from scratch and backtests it across 28 years of daily NEPSE data, from July 1997 to March 2025.

The strategy

The system is long-only and relies on three signals. A Z-score filter identifies short-term price deviations from recent history, functioning as a mean-reversion trigger. The 14-period Relative Strength Index detects oversold conditions. And a 240-day simple moving average acts as a trend regime filter, ensuring that mean-reversion entries only occur when the broader market direction is upward.

Buy signals require all three conditions to align. Risk management is handled through dynamic position sizing, a maximum of four concurrent positions, cooldown periods between trades, and a fixed risk-reward ratio of 1:4.5. A flat 0.25% execution adjustment is applied to all exits as a simplified friction estimate.

This is not a complex system. The components are textbook indicators available in any technical analysis library. The contribution lies in testing them in a market where nobody had tested them before.

The numbers

Over the full sample, the strategy produced a CAGR of 14.38% compared to 11.47% for Buy and Hold. On a risk-adjusted basis, the gap widens considerably: a Sharpe ratio of 1.39 versus 0.64, and a Sortino ratio of 1.35 versus 0.89. Maximum drawdown was 28.11% for the quant strategy against 75.16% for buy and hold (Poudel and Paudel, 2025).

The drawdown comparison is the most striking result. NEPSE experienced a severe bear market around 2012 that drew the index down by roughly three quarters from its peak. The quantitative strategy, by moving to cash when the price sat below its 240-day average, avoided most of that decline. Capital preservation in a market that offered no hedging instruments and no ability to go short was the main advantage.

The strategy was invested only 49.5% of trading days. That selectivity improved risk-adjusted returns but came at a cost: during several multi-year stretches (2002-2004, 2010-2012, 2018-2019, 2023-2024), capital sat idle in cash, earning nothing. The authors acknowledge this opportunity cost but do not quantify it against a risk-free rate.

A Monte Carlo simulation with 20,000 randomised trade sequences produced no bankruptcies and no negative final outcomes. The average ending capital was roughly NPR 11 million from a starting base that generated about NPR 29 million in the actual backtest. The worst-case scenario still ended in profit. Average maximum drawdown across all simulations was 10.2%, with the worst single path reaching 27.2%.

What the sensitivity analysis actually shows

The authors run two parameter perturbation tests. The first makes modest changes: RSI thresholds shift from 35/60 to 30/65, the moving average changes from 240 to 200 days. Results barely move. CAGR drops from 14.38% to 14.01%, Sharpe from 1.39 to 1.32. This suggests the strategy is not tightly fitted to a single parameter combination within that neighbourhood.

The second test is more aggressive: longer cooldown periods, wider stop-losses, a longer Z-score lookback window. Here the degradation is real. CAGR falls to 9.91%, Sharpe drops to 1.05, and total return collapses from 2,877% to 987%. The profit factor remains above 1.5, so the strategy stays profitable, but the magnitude of change should not be glossed over. A factor of three in total return from parameter shifts is not what robust performance looks like.

The honest interpretation: the strategy works across a range of parameters, but that range has boundaries, and crossing them costs more than the authors emphasise.

Statistical validation and its limits

A Kolmogorov-Smirnov test confirms that the return distributions of the two strategies differ significantly (p < 0.001). A Z-test on Sharpe ratios yields a statistic of 4.01 with p < 0.001, rejecting the null that the quant strategy's risk-adjusted returns are equal to buy and hold (Poudel and Paudel, 2025).

These tests answer the question they are designed for. But they do not address selection bias, data-snooping, or the fact that the same dataset was used for both strategy development and evaluation. There is no walk-forward test, no out-of-sample period, and no combinatorial purged cross-validation. The Monte Carlo approach shuffles the sequence of existing trades rather than generating truly independent scenarios. For a strategy this simple, overfitting risk is lower than for a machine learning model, but it is not zero.

The paper is published in the Quest Journal of Management and Social Sciences, a Nepalese academic journal. The research context matters: this is an early-stage contribution to a local academic discourse where algorithmic trading has received almost no attention. Judging it by the standards of a Journal of Financial Economics submission would miss the point.

What transfers and what does not

The transferable lesson is about market structure, not about the specific parameter values. NEPSE is a market where behavioural biases are extreme, institutional participation is minimal, and basic risk management is rare. In that environment, even simple rules that enforce trend alignment, entry discipline, and position limits produce measurably better outcomes than the alternative of holding through 75% drawdowns.

This aligns with findings from other emerging markets. Chaudhuri and Wu (2003) documented mean reversion in stock prices across 17 emerging markets. Butt, Kolari, and Sadaqat (2021) confirmed momentum effects in 19 emerging markets, while noting that returns are weaker and more fragile than in developed economies. The NEPSE results fit this pattern: the edge exists, but it is smaller and more sensitive to implementation details than developed-market backtests typically suggest.

What does not transfer is the assumption that these specific indicators and thresholds will work elsewhere. The 240-day moving average filter works well on a market that produces long, grinding bear phases lasting two to four years. A market with faster regime changes would need different parameters. The absence of short selling makes the long-only constraint realistic here but irrelevant in markets where hedging is available. And the 0.25% execution adjustment, while possibly reasonable for a low-frequency strategy on a single index, would be inadequate for most practical implementations involving individual stocks.

What this means for systematic investing

The study adds a data point to a growing body of evidence that rule-based approaches outperform discretionary trading in markets where behavioural noise is highest. The mechanism is not prediction. It is discipline: avoiding large drawdowns by staying out of unfavourable regimes, sizing positions relative to risk, and removing the temptation to chase trends that have already played out.

The more interesting observation is about market development. NEPSE currently operates without short selling, with limited electronic infrastructure, and with almost no quantitative research tradition. As markets like these evolve, the adoption of even basic systematic methods could reduce the severity of boom-bust cycles driven by retail herding. Whether that happens through academic influence, institutional adoption, or regulatory reform remains an open question.

For systematic investors operating in more developed markets, the paper is a useful reminder that the core advantage of rules-based allocation is not sophistication. It is consistency. The simplest possible trend filter, applied with discipline and basic risk controls, captured most of the benefit in this study. Adding complexity did not improve the result; changing parameters too aggressively degraded it.

The edge is in the process, not in the indicator.

References:

Butt, H.A., Kolari, J.W. and Sadaqat, M. (2021) 'Revisiting momentum profits in emerging markets', Pacific Basin Finance Journal, 65, 101455. doi: 10.1016/j.pacfin.2020.101455.

Chaudhuri, K. and Wu, Y. (2003) 'Mean reversion in stock prices: Evidence from emerging markets', Managerial Finance, 29(10), pp. 22-37. doi: 10.1108/03074350310768490.

Poudel, P. and Paudel, S. (2025) 'Quantitative Trading Strategy, Backtesting, and Performance Analysis Using Python: A Data-Driven Analysis', Quest Journal of Management and Social Sciences, 7(2), pp. 219-238. doi: 10.3126/qjmss.v7i2.87782.

Disclaimer: Educational and informational purposes only. Past performance does not guarantee future results. Not investment advice.