Investigation Report · April 2026 · Complete

AI-Powered NEPSE Stock Prediction Platform

Combining five years of NEPSE historical data, NRB macroeconomic indicators, and NLP-driven Nepali financial news sentiment to deliver explainable Buy, Hold, and Sell signals for retail investors.

Live Market Data Research

Gaurab Nepali · NP069821 BSc (Hons) IT · APU/LBEF CT052-3-3-IR

NEPSE INDEX · LIVE LIVE DATA

2,786.35

+27.86+1.01%

TURNOVER

NPR 4.49B

TRADES

85,879

SCRIPS

242

LIVE MARKET DATA

NEPSE Listed Stocks

Fetching live NEPSE data…

01 · THE PROBLEM

Why NEPSE needs intelligent analytics

PROBLEM 01

Technical analysis ignores macro reality

RSI, Bollinger Bands and moving averages can't capture NRB rate changes that move the market.

0.5% NEPSE per 0.1% inflation rise

PROBLEM 02

No integrated analytical platform

Nothing combines NEPSE prices, NRB macro data and Nepali news sentiment for retail investors.

PROBLEM 03

Structural information asymmetry

85% of participants are retail investors without institutional tools; ~60% lose money.

60% of retail investors lose money

PROJECT VISION

An ML platform built for Nepal's market

Linear Regression models macro correlations while a Decision Tree captures non-linear reactions. An NLTK sentiment module adds 10–12% accuracy, producing clear Buy/Hold/Sell signals — aligned with UN SDG 8, Target 8.10.

Prof. (Dr.) R.N. Thakur

Primary Supervisor · Dean, LBEF

Mr. Ramesh Suwal

Co-Supervisor · System Development

UN SDG 8 · Target 8.10 · Financial Inclusion

02 · LITERATURE REVIEW

Findings that validate this project

Macro factors drive NEPSE

Inflation, rates and GDP are validated predictors — macro inclusion raises accuracy ~15%.

+15% accuracy

NLP sentiment helps

Positive Nepali news lifts the index ~0.11% next day; NLP adds 8–12% in emerging markets.

+0.11% next-day

Explainability builds trust

59.4% trust ML only with explanations — Decision Trees give human-readable rules.

Bloomberg costs $24k/yr

No tool covers NEPSE data, retail access, macro and Nepali sentiment together.

Lightweight ML fits Nepal

Limited bandwidth and devices make LSTM/Transformer impractical.

NEPSE lacks analytics

The official portal shows only raw prices — no predictions or sentiment.

03 · TECHNOLOGY STACK

Every tool chosen with evidence

Python 3.11

LANGUAGE

Scikit-learn, Pandas, NLTK ecosystem.

Scikit-learn

ML FRAMEWORK

Linear Regression + Decision Tree.

NLTK

NLP / SENTIMENT

Nepali financial lexicon scoring.

BeautifulSoup · pdfplumber

DATA COLLECTION

News scraping + NRB PDF extraction.

React.js 18

FRONTEND

Responsive dashboard (62.5% mobile).

Flask 3.0

BACKEND API

REST API + SQLAlchemy.

PostgreSQL 16

DATABASE

Time-series + JSONB sentiment.

AWS · Docker · Redis

DEPLOYMENT

Containerized EC2 + caching.

04 · PRIMARY RESEARCH

32 investors surveyed — the data speaks

87.5%

Want Buy/Hold/Sell signals

Most-requested feature — confirms UR-01 as essential MVP.

68.8%

Dissatisfied with current tools

Can't predict movement after macro announcements.

62.5%

Primarily use smartphones

Confirms UR-08: must be mobile-responsive.

78.1%

Use only basic indicators

No access to integrated sentiment.

59.4%

Trust AI with explanation

Confirms Decision Tree selection and UR-07.

Validated requirements

From survey, interviews, literature and PSF.

05 · RESOURCES

Data sources & tools