Time-series Analysis : Nifty-50 stock price dataset

Time series analysis enables us to visualize trends in stock prices, seasonality, autocorrelations and hence predict subsequent prices and trends. These insights are essential for many financial and non financial companies, portfolio analysis etc. In this notebook, I have explored various feature engineering techniques and models exclusive to time series data.
Please find the link to my kaggle notebook here: https://www.kaggle.com/breenda/lstm-rnn-arimax-ar-models-and-analysis

---

Goal of this project was to explore the techniques to handle stock data. I started of with visualizing the various parameters to get an idea of the trends in data, seasonality, if its present etc. Next, I explored a few engineering techniques to generate exogenous features. Then I used models such as ARIMA, Autoregressors, LSTMs and RNNs to see which models best predict the target variable.

Dataset

https://www.kaggle.com/rohanrao/nifty50-stock-market-data

Libraries Used:

Statsmodels

Keras

Numpy

Pandas

Sci-kit learn

Sci-py

pmdarima

Matplotlib

Seaborn

Features:

1. Series: Here EQ stands for equity series of stock market.
2. Prev Close: The closing price of the stock for the day before.
3. Open,High, Low, Last, Close: The opening price, highest price, lowest price, last price and closing price of ICICI shares on the current day.
4. VWAP: Volume Weighted Average Price,the target variable to predict. VWAP is a trading benchmark used by traders that gives the average price the stock has traded at throughout the day, based on both volume and price.
5. Volume: Volume of shares traded on the current day.
6. Turnover: It is a measure of stock liquidity calculated by dividing the total number of shares traded over a period by the average number of shares outstanding for the period.
7. Trades: total number of trades on the current day.
8. Deliverable Volume: is the quantity of shares which actually move from one set of people to another set of people.
9. Deliverable(%): Deliverable volume in percentage.

Feature analysis:

Trends in VWAP, Trades and Turnovers over a period of 20 years

Box Cox Plot of VWAP

Exogenous features generated by moving average smoothing and rolling of features

Auto Correlation plots

Stationarity of the data by Augmented Dickey-Fuller test

Decomposition to study trends and seasonality in VWAP

Models explored:

Statsmodels Autoregressors: rmse score - 32.30

pmdarima ARIMAX: rmse score - 82.27 mae score - 55.83

keras LSTM sequential model: rmse score - 0.046 mae score- 0.0370

keras RNN sequential model: rmse score - 0.046 mae score- 0.0624