An ML-powered NEPSE prediction system combining historical data, NRB macroeconomic indicators, and NLP sentiment analysis to deliver clear Buy, Hold & Sell signals.
Three interrelated, validated problems undermine NEPSE retail investor outcomes — none addressed by any existing platform.
Basic indicators — Bollinger Bands, RSI, moving averages — are insufficient for Nepal's macro-sensitive market. Each 0.1% rise in inflation causes a 0.5% drop in the NEPSE index, yet one-dimensional tools cannot capture this.
Systematic signal errorsNo platform combines NEPSE historical prices, NRB macroeconomic data, and Nepali financial news sentiment. Incorporating sentiment analysis can improve forecasting accuracy by 12% in emerging markets.
Zero integrated solutionsInstitutional investors use sophisticated proprietary systems. Retail investors rely on social networks and informal information channels, creating a structural disadvantage that directly contributes to the 60% loss rate.
Structural inequalityThe proposed system integrates data collection, ML prediction, sentiment analysis, and a user-facing dashboard into a cohesive, Nepal-optimised platform.
Automated scraping of daily news from Sharesansar, Himalayan Times and Kathmandu Post using BeautifulSoup. pdfplumber extracts NRB quarterly macroeconomic PDFs. ETL pipeline feeds PostgreSQL with Z-score outlier detection and min-max normalisation.
Linear Regression models linear correlations between macroeconomic indicators and stock prices. Decision Tree Classifier captures non-linear market event responses. Both models are combined with sentiment scores to produce final Buy, Hold, or Sell signals.
NLTK-based keyword sentiment classifier assigns positive, negative, or neutral scores to Nepali financial news articles. Domain-specific financial lexicon calibrated for NEPSE terminology. Sentiment scores integrated as a feature input to the ML pipeline.
React.js 18 with Chart.js delivers mobile-responsive, interactive price charts, sentiment trend graphs and sector-wise comparisons. Flask 3.0 REST API backend with Redis caching ensures sub-3-second response on low-bandwidth connections. Secure JWT authentication.
Every technology choice was rigorously justified against six criteria including interpretability, infrastructure fit, and retail investor trust.
No existing platform simultaneously addresses NEPSE data integration, retail accessibility, macroeconomic data, and Nepali news sentiment.
| Criterion | Bloomberg Terminal | QuantConnect | Sensibull | NEPSE Portal | Proposed System |
|---|---|---|---|---|---|
| NEPSE data support | No | No | No | Basic | Yes (full) |
| Retail accessibility | No | Limited | Yes | Yes | Yes |
| Sentiment analysis | Yes | Limited | No | No | Yes (NLP) |
| Macroeconomic integration | Yes | Limited | No | No | Yes (NRB) |
| ML-based predictions | Yes | Yes | Partial | No | Yes |
| Buy / Hold / Sell signals | Yes | Yes | Yes | No | Yes |
| Cost to end user | $24,000/yr | Free/paid | Free/paid | Free | Free |
| Nepal infrastructure fit | No | No | No | Yes | Yes |
The proposed system is the only platform satisfying all eight criteria simultaneously.
Agile was selected for continuous user feedback integration, iterative ML development support, and adaptability to evolving requirements.
System architecture, database schema, Docker containerisation, GitHub repository structure, CI/CD pipeline via GitHub Actions. Full product backlog compiled from 14 user requirements.
BeautifulSoup web scraping, pdfplumber NRB PDF extraction, ETL pipeline to PostgreSQL. Z-score outlier detection and min-max normalisation validated on sample datasets.
Linear Regression and Decision Tree trained with 80/20 split on 2020–2024 data. Evaluated with MAE, RMSE, and Directional Accuracy against moving average and Bollinger Band baselines.
NLTK keyword sentiment analysis with domain-specific Nepali financial lexicon. Calibrated for positive/negative/neutral classification, integrated as feature input to the prediction pipeline.
Flask REST API, React.js dashboard with Chart.js interactive charts, sentiment graphs, and macroeconomic correlation visualisations. Redis caching for real-time data access optimisation.
Full system integration, AWS EC2 cloud deployment, and User Acceptance Testing with 20–30 retail investor pilot participants from NEPSE brokerage communities using the System Usability Scale (SUS).
Mixed-methods research — 32 verified NEPSE retail investor surveys plus 5 qualitative expert interviews — grounded every user requirement in empirical evidence.
Key quantitative findings (n=32)
Qualitative interview themes (n=5)
Frustration with time lag between macroeconomic announcements and availability in existing trading tools. Directly informed UR-03, UR-04, and UR-05 — push notifications for significant market events.
Experienced investors require clear confidence levels to trust algorithmic recommendations. This theme was the primary driver behind UR-07 (display model accuracy metrics) and the selection of transparent Decision Tree models over black-box deep learning.
Significant variance in quantitative financial literacy across participants. Led to the visual-first, simplified signals language design strategy and the requirement to provide explanations alongside every recommendation.
Participants
Derived from quantitative survey analysis, qualitative interviews, domain literature review, and PSF documentation.
| ID | Requirement | Priority | Source |
|---|---|---|---|
| UR-01 | Provide Buy, Hold, or Sell signals for NEPSE stocks | Essential | Survey, Interview |
| UR-02 | Display interactive price trend charts with historical data | Essential | Survey |
| UR-03 | Integrate NRB macroeconomic data into predictions | High | Literature, Interview |
| UR-04 | Provide daily news sentiment scores with source citation | High | Survey, Interview |
| UR-05 | Deliver push notifications for significant market events | High | Survey |
| UR-06 | Allow filtering of predictions by market sector | Medium | Survey |
| UR-07 | Display model accuracy metrics (MAE, RMSE) to users | High | Interview |
| UR-08 | Provide mobile-responsive interface design | Essential | Survey |
| UR-09 | Require secure user registration and login | Essential | Survey |
| UR-10 | Allow admin to retrain ML models on updated datasets | Medium | PSF Document |
| UR-11 | Support English interface with Nepali annotations | Medium | Survey |
| UR-12 | Operate acceptably on low-bandwidth connections | High | Literature, PSF |
| UR-13 | Provide downloadable prediction summary reports | Low | Survey |
| UR-14 | Admin can view user activity logs and system usage statistics | Medium | PSF Document |
The system advances both practical investor outcomes and broader societal goals aligned with the UN Sustainable Development Goals.
Web-based Buy/Hold/Sell signal generation with NEPSE stock price trend charts and volatility indexes accessible to every retail investor.
Ingests NEPSE prices/volumes, NRB quarterly PDFs, and Nepali finance media continuously — no manual data entry required.
10–12% improvement in prediction accuracy through positive/negative/neutral classification of Nepali financial news.
Equalises access to sophisticated analytical tools between institutional traders and retail investors — directly addressing structural inequality.
Improved analytical access for the 85% retail investor majority, building evidence-based confidence and reducing loss rates.
Directly contributes to SDG 8 Target 8.10 — Decent Work and Economic Growth through improved financial market access.
Supervisors
This investigation was undertaken to address a critical gap in Nepal's financial technology landscape. With over 85% of NEPSE investors being retail participants — most lacking access to any sophisticated analytical platform — the potential for a data-driven, ML-powered system to meaningfully reduce the 60% loss rate among retail investors is significant.
The investigation combined a rigorous literature review, comparative analysis of four global systems, and original mixed-methods primary research (32 surveys + 5 expert interviews) conducted across Kathmandu's NEPSE brokerage communities between February and April 2026.
Five mandatory supervisory sessions with Prof. (Dr.) R.N. Thakur guided each critical stage: PSF approval, literature review depth, methodology design, critical analysis standards, and final submission readiness.