The S&P BSE SENSEX is a free-float market-weighted stock market index of 30 well-established and financially sound companies listed on Bombay Stock Exchange. In this project, I shall analyze historical S&P BSE Sensex data, particularly the Open, High, Low and Close over the past 10 years. I shall then calculate various technical indicators used in market analysis to forecast BSE market performance, using the famous XGBoost regressor. Then, I shall do the same using LSTMs, compare the outcomes and finalize my model.

This project has 5 steps:

1. Data Preprocessing
   This dataset is clean and requires no preprocessing.

df = pd.read_csv(Location of dataset, index_col=False)
dfp=df
df['Date']=df['Date'].astype('datetime64')
df.head()

2. Data Visualization
   Candlestick charts are the language of stocks. Also, since we will be using time series forecasting, it only makes sense to decompose the data and evaluate trends and seasonality. The code can be found in dataviz.py.

3. Plotting Technical Indicators
   Technical indicators are the stock trader's toolbox. I start off by plotting exponential and simple moving averages over different time periods. Then, I plot the RSI, which is an indicator of whether the stocks of the index are oversold or overbought. Finally, I plot the MACD to observe the market momentum. The code can be found in techind.py.

4. Using XGBoost to predict closing prices
   Not all indicators are equally important. Understanding the weightage alloted to each one allows us to make optimal decisions. We shall use GridSearch to find the parameters with the best validation score and use those in our model. The code can be found in xgb.py.

5. Using LSTMs for time series forecasting
   Our model doesn't do a very good job of predictions. Also, it relies on calculation of a lot of parameters and is computationally expensive. A better option would be to use LSTMs for predicting the future closing prices. The code can be found in lst.py.

BSE allows you to download historical data (Open, High, Low, Close) of the indices. You can find the link to do so here: https://www.bseindia.com/indices/IndexArchiveData.html

• Python 3.6
• Numpy
• Pandas
• Xgboost
• Sklearn
• Plotly
• Stldecompose

#importing relevant libraries
import os
import numpy as np
import pandas as pd
import xgboost as xgb
import matplotlib.pyplot as plt
from xgboost import plot_importance, plot_tree
from sklearn.metrics import mean_squared_error
from sklearn.preprocessing import MinMaxScaler
from sklearn.model_selection import train_test_split, GridSearchCV
import types
from botocore.client import Config
import ibm_boto3

# Plotting    
!pip install plotly==4.6.0
import plotly as py
import plotly.io as pio
import plotly.graph_objects as go
from plotly.subplots import make_subplots
from plotly.offline import download_plotlyjs, init_notebook_mode, plot, iplot

# Time series decomposition
!pip install stldecompose
from stldecompose import decompose

You can find the presentation on YouTube
If you wish to view the interactive plots, check out this project on IBM Watson Studio

1. Decomposition : Barring recent stock market changes because of the CoVID-19 pandemic, we observe that Sensex seems to have a strong seasonality and limited noice/residulaity. In simpler words, this means that stocks markets behave in a predicatable manner over the long term.

2. Feature Importance: Of the listed technical indicators, 9 day Exponential Moving Average and Relative Strength Index are the most important ones.

3. LSTM Predictions: The LSTM model does a really good job of predicting future market trends, due to the high seasonality and predictable noise of Sensex.

Akshat Kharbanda is a BITS Pilani, KK Birla Goa Campus Student majoring in Electronics and Communication Engineering. Feel free to connect on LinkedIn!