Bank Nifty Predictive Analysis through

Time Series Forecasting

Submitted By: Supervised by:

Mohd Saad Khan Prof Dr. Moonis Shakeel
Roll No.21MFE019 Department of Economics
Mohd Owais Jamia Millia Islamia
Roll no. 21MFE016

Course: MSc. Banking and Financial analytics

Sem- 3rd
Introduction
● Bank Nifty prediction using machine learning helps you discover the future value of banking
stock and other financial asset trading on a exchange.
● The entire idea of predicting bank nifty is to gain significant profit.
● Machine learning models such as LSTM’s, ARIMA and GARCH are popular models applied to
predicting time series data of Stock market.
● It is vital to examine the accuracy of forecasts because they have several restrictions in
application
● The major goal of this paper is to see which forecasting approaches provide the best
prediction in terms of minimum squared error and maximum forecasting accuracy.
Research Methodologies
Data Collection
● We’ve imported our data through google sources like yahoo finance.
Data and Pattern analysis
● First we have sorted the data through panda library. Then we’ve plot the data through
libraries like seaborn and matplotlib.
Popular forecasting methods
● Methods we’ll use to predict the nifty bank stock prices are LSTM’s ,ARIMA, and GARCH.
Assessment Metric
● Root-mean-square error commonly used metric for evaluating the accuracy of a model’s
prediction. There is also other metrics to evaluate the accuracy of the model which are F
score, P-score and Classification report etc.
Internship experience

● Account opening
● Checque clearing
● Locker Notices
● Updating database manually
● E- KYC
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
plt.style.use('fivethirtyeight')
import yfinance as yf
import math
from sklearn.preprocessing import MinMaxScaler

asss=['AUBANK','AXISBANK','BANDHANBNK','BANKBARODA','FEDERALBNK','HDFCBANK','ICICIBANK',
components=list(asss)

components

['AUBANK',
'AXISBANK',
'BANDHANBNK',
'BANKBARODA',
'FEDERALBNK',
'HDFCBANK',
'ICICIBANK',
'IDFCFIRSTB',
'INDUSINDBK',
'KOTAKBANK',
'PNB',
'SBIN']

for i in range(len(components)):
components[i]=components[i]+'.NS'
components

['AUBANK.NS',
'AXISBANK.NS',
'BANDHANBNK.NS',
'BANKBARODA.NS',
'FEDERALBNK.NS',
'HDFCBANK.NS',
'ICICIBANK.NS',
'IDFCFIRSTB.NS',
'INDUSINDBK.NS',
'KOTAKBANK.NS',
'PNB.NS',
'SBIN.NS']

components.append('^NSEBANK')
components

['AUBANK.NS',
'AXISBANK.NS',
'BANDHANBNK.NS',
'BANKBARODA.NS',
'FEDERALBNK.NS',
'HDFCBANK.NS',
 'HDFCBANK.NS',
'ICICIBANK.NS',
'IDFCFIRSTB.NS',
'INDUSINDBK.NS',
'KOTAKBANK.NS',
'PNB.NS',
'SBIN.NS',
'^NSEBANK']

data=yf.download(components,start='2007-4-1',end='2023-03-31')
data

[*********************100%***********************] 13 of 13 completed
Adj Close

AUBANK.NS AXISBANK.NS BANDHANBNK.NS BANKBARODA.NS FEDERALBNK.NS HDFCBANK

Date

2007-04-02 NaN 81.345421 NaN 31.156727 9.322885

2007-04-03 NaN 82.443481 NaN 31.257248 9.465549

2007-04-04 NaN 81.239159 NaN 31.589773 9.490828

2007-04-05 NaN 79.955132 NaN 33.066803 9.992247

2007-04-09 NaN 88.438591 NaN 34.837681 10.341372

... ... ... ... ... ...

2023-03-23 569.650024 848.799988 212.500000 162.699997 127.000000 1563.150

2023-03-24 559.900024 839.900024 203.100006 159.800003 126.849998 1560.650

2023-03-27 566.900024 833.349976 197.800003 161.250000 125.449997 1567.449

2023-03-28 563.549988 832.450012 187.449997 160.750000 125.699997 1580.199

2023-03-29 585.549988 842.599976 189.899994 164.449997 129.100006 1587.800

3947 rows × 78 columns

banknifty=yf.download('^NSEBANK',start='2007-4-1',end='2023-4-1')

[*********************100%***********************] 1 of 1 completed

banknifty

Open High Low Close Adj Close Volume

Date

2007-09-17 6898.000000 6977.200195 6843.000000 6897.100098 6897.020020

 2007-09-18 6921.149902 7078.950195 6883.600098 7059.649902 7059.567871

2007-09-19 7111.000000 7419.350098 7111.000000 7401.850098 7401.764160

2007-09-20 7404.950195 7462.899902 7343.600098 7390.149902 7390.063965

2007-09-21 7378.299805 7506.350098 7367.149902 7464.500000 7464.413086

... ... ... ... ... ...

2023-03-24 39555.250000 39767.898438 39294.898438 39395.351562 39395.351562 166100

2023-03-27 39484.699219 39695.199219 39273.750000 39431.300781 39431.300781 194200

2023-03-28 39545.050781 39645.199219 39326.101562 39567.898438 39567.898438 202300

2023-03-29 39611.550781 40055.000000 39609.550781 39910.148438 39910.148438 259600

2023-03-31 40231.250000 40690.398438 40180.199219 40608.648438 40608.648438 188000

3538 rows × 6 columns

from datetime import date

plt.figure(figsize=(16,8))
plt.title('Price Chart')
plt.plot(banknifty['Close'])
plt.xlabel('Date', fontsize=18)
plt.ylabel('Closing Price',fontsize=18)
plt.show()
## volume graph
banknifty['Volume'].plot(label='nifty bank volume',figsize=(16,8))

<Axes: xlabel='Date'>
### log return
logreturn=banknifty['Log Return']=np.log(banknifty['Close']/banknifty['Close'].shift(1))
logreturn

Date
2007-09-17 NaN
2007-09-18 0.023294
2007-09-19 0.047335
2007-09-20 -0.001582
2007-09-21 0.010010
...
2023-03-24 -0.005608
2023-03-27 0.000912
2023-03-28 0.003458
2023-03-29 0.008612
2023-03-31 0.017350
Name: Close, Length: 3538, dtype: float64

banknifty['Log Return'].plot(title='log return',figsize=(16,8))

<Axes: title={'center': 'log return'}, xlabel='Date'>

# histogram adjusted close
plt.hist(banknifty['Close'],bins=500)
plt.show

<function matplotlib.pyplot.show(close=None, block=None)>

## histogram volume
plt.hist(banknifty['Volume'])
plt.show()
#only closing data
close_df=banknifty.filter(['Close'])

#convert to numpy array

dataset=close_df.values

dataset

array([[ 6897.10009766],
[ 7059.64990234],
[ 7401.85009766],
...,
[39567.8984375 ],
[39910.1484375 ],
[40608.6484375 ]])

#size of training data

training_data_len=math.ceil(len(dataset)*0.8)
training_data_len

2831

#scaling the data

scaler= MinMaxScaler(feature_range=(0,1))
scaled_data=scaler.fit_transform(dataset)

scaled_data

array([[0.08738522],
[0.09137815],
[0.09978408],
...,
[0.88992216],
[0.89832931],
[0.9154875 ]])

training_dataset= scaled_data[0:training_data_len,:]
#spliitting into xtrain and ytrain
x_train=[]
y_train=[]
for i in range(60,len(training_dataset)):
x_train.append(training_dataset[i-60:i,0])
y_train.append(training_dataset[i,0])
if i<=61:
print(x_train)
print(y_train)
print()

[array([0.08738522, 0.09137815, 0.09978408, 0.09949667, 0.10132303,

0.10590183, 0.10536755, 0.10848108, 0.11039096, 0.11551386,
0.11417019, 0.11688209, 0.1153591 , 0.11067591, 0.1053 ,
0.11191887, 0.11523014, 0.11837806, 0.112856 , 0.12151001,
0.12558525, 0.11693122, 0.10486644, 0.10032326, 0.10386544,
0.1169607 , 0.12009389, 0.12500676, 0.13289806, 0.13906002,
0.13742404, 0.13936462, 0.14392006, 0.15281483, 0.14880715,
0.14435486, 0.1412573 , 0.13600175, 0.13310194, 0.14051545,
0.15410076, 0.15021101, 0.15068879, 0.14877522, 0.14586679,
0.13486074, 0.13901211, 0.13849873, 0.1420495 , 0.14077952,
0.13832553, 0.14209741, 0.14827165, 0.14880715, 0.14805916,
0.15332084, 0.15427518, 0.15859726, 0.16027992, 0.16583393])]
[0.16499874668438996]

[array([0.08738522, 0.09137815, 0.09978408, 0.09949667, 0.10132303,

0.10590183, 0.10536755, 0.10848108, 0.11039096, 0.11551386,
0.11417019, 0.11688209, 0.1153591 , 0.11067591, 0.1053 ,
0.11191887, 0.11523014, 0.11837806, 0.112856 , 0.12151001,
0.12558525, 0.11693122, 0.10486644, 0.10032326, 0.10386544,
0.1169607 , 0.12009389, 0.12500676, 0.13289806, 0.13906002,
0.13742404, 0.13936462, 0.14392006, 0.15281483, 0.14880715,
0.14435486, 0.1412573 , 0.13600175, 0.13310194, 0.14051545,
0.15410076, 0.15021101, 0.15068879, 0.14877522, 0.14586679,
0.13486074, 0.13901211, 0.13849873, 0.1420495 , 0.14077952,
0.13832553, 0.14209741, 0.14827165, 0.14880715, 0.14805916,
0.15332084, 0.15427518, 0.15859726, 0.16027992, 0.16583393]), array([0.09137815, 0
0.10536755, 0.10848108, 0.11039096, 0.11551386, 0.11417019,
0.11688209, 0.1153591 , 0.11067591, 0.1053 , 0.11191887,
0.11523014, 0.11837806, 0.112856 , 0.12151001, 0.12558525,
0.11693122, 0.10486644, 0.10032326, 0.10386544, 0.1169607 ,
0.12009389, 0.12500676, 0.13289806, 0.13906002, 0.13742404,
0.13936462, 0.14392006, 0.15281483, 0.14880715, 0.14435486,
0.1412573 , 0.13600175, 0.13310194, 0.14051545, 0.15410076,
0.15021101, 0.15068879, 0.14877522, 0.14586679, 0.13486074,
0.13901211, 0.13849873, 0.1420495 , 0.14077952, 0.13832553,
0.14209741, 0.14827165, 0.14880715, 0.14805916, 0.15332084,
0.15427518, 0.15859726, 0.16027992, 0.16583393, 0.16499875])]
[0.16499874668438996, 0.16165185396165127]

## convering to numpy arrays

x_train=np.array(x_train)
y_train=np.array(y_train)
## lstm works with only 3d data

## reshaping data
x_train=np.reshape(x_train,(x_train.shape[0],x_train.shape[1],1))

x_train.shape

(2771, 60, 1)

from tensorflow.keras.layers import Dense, Dropout,LSTM

from tensorflow.keras.models import Sequential

## building LSTM
model=Sequential()
model.add(LSTM(50,return_sequences=True,input_shape=(x_train.shape[1],1)))
model.add(LSTM(50,return_sequences=False))
model.add(Dense(25))
model.add(Dense(1))

##compiling model
model.compile(optimizer='adam',loss='mean_squared_error')

model.fit( x_train, y_train, batch_size=1, epochs=1)

2771/2771 [==============================] - 68s 23ms/step - loss: 8.4483e-04

<keras.callbacks.History at 0x7fc98f73c610>

##create the testing data set

## creating neew array containing scaled values from 2771 to 3528
testing_data=scaled_data[training_data_len-60: , :]
#create the data set
x_test=[]
y_test=dataset[training_data_len:, :]
for i in range(60, len(testing_data)):
x_test.append(testing_data[i-60:i,0])

## converting to numpy array

x_test=np.array(x_test)

##reshaping the data

x_test=np.reshape(x_test,(x_test.shape[0],x_test.shape[1],1))

## get the model's predicted price values

prediction=model.predict(x_test)
prediction=scaler.inverse_transform(prediction)

23/23 [==============================] - 1s 13ms/step

## model evaluation
## get root meaned squared error(RMSE)
rmse_model=np.sqrt( np.mean(prediction- y_test)**2)
rmse_model

1417.3565986341937

## plot the data

train=close_df[:training_data_len]
valid=close_df[training_data_len:]
valid['prediction']=prediction
## visualization
plt.figure(figsize=(16,8))
plt.title('Model output')
plt.xlabel('Date',fontsize=18)
plt.ylabel('Close price INR',fontsize=18)
plt.plot(train['Close'])
plt.plot(valid[['Close','prediction']])
plt.legend(['Train','Val','Predictions'],loc='lower right')
plt.show()

<ipython-input-45-fdc1f080e610>:4: SettingWithCopyWarning:
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_g

valid['prediction']=prediction
## actual and predicted prices
valid

Close prediction

Date

2020-05-29 19297.250000 17952.277344

2020-06-01 19959.900391 18246.832031

2020-06-02 20530.199219 18636.347656

2020-06-03 20940.699219 19099.777344

2020-06-04 20390.449219 19595.921875

... ... ...

2023-03-24 39395.351562 37641.164062

2023-03-27 39431.300781 37667.566406

2023-03-28 39567.898438 37673.480469

2023-03-29 39910.148438 37684.117188

2023-03-31 40608.648438 37740.035156

707 rows × 2 columns