The Math of Movement: Defining the ±1% Threshold

In financial forecasting, the most common instinct is to predict an exact price (regression). However, for the Nepal Stock Exchange (NEPSE), price regression often leads to high error rates and overfitted models that fail to capture tradeable signals. This research shifts the focus from “How much will it cost?” to “Which way is it moving?”.

1. Why Classification Beats Regression

Predicting that a stock like SHIVM will be exactly Rs. 542.30 tomorrow is statistically improbable. For a researcher, the more valuable question is whether the stock will move significantly enough to warrant a position. By treating this as a multi-class classification task, we filter out the “market noise” and focus the model on high-conviction movements.

2. The Mathematical Framework

We define the price movement of the next trading day ($t+1$) based on a percentage return threshold. The logic is encapsulated in the following piecewise function:

3. The “Goldilocks” Threshold: Why 1%?

Selecting the threshold is a balancing act between sensitivity and signal reliability. My analysis of 221,473 records from March 2024 to January 2026 compared several boundaries:

4. Filtering the Noise

By setting the threshold at ±1%, we discover that roughly 34% of trading days in the NEPSE show significant movement. The remaining 66% (the Neutral class) represents the “sideways” movement where deep learning models often struggle to find patterns.

This 1% barrier ensures that the LSTM, BiLSTM, and Attention-based models I am testing are not wasting computational energy on negligible fluctuations. Instead, they are trained to identify the specific technical indicators—like RSI, MACD, and Bollinger Bands—that precede a meaningful breakout or breakdown.

5. Research Implications

Existing NEPSE studies have largely focused on index-level movements rather than individual high-liquidity stocks. By applying this ±1% threshold to the Top 30 stocks by turnover (including HRL, NRN, and NRIC), this research aims to provide the first comprehensive direction classification study in the Nepalese context.