markets-insights

Markets Insights 📝

This package fetches and processes capital markets data from NSE (National Stock Exchange, India). Following data can be retrieved

1. Index (Nifty, Bank Nifty, NiftyIT)
2. Stocks
3. Derivatives (Futures and Options)

Support for additional markets and instruments can be added externally

The package can perform technical functions on price of Index and Stocks. Following indicators are supported.

1. Simple Moving Averages (SMA)
2. Relative Strength Index (RSI)
3. Stochastic RSI
4. Bollinger Bands (with standard deviations)

The calculation pipeline is quite extensible and more idicators can be added externally.

Getting Started 🚀

Installation

!pip install markets_insights

from markets_insights.core.environment import Environment
Environment.setup(cache_data_base_path = './cache-data')

Get Index data for date range

from markets_insights.datareader import data_reader
import datetime

reader = data_reader.NseIndicesReader()

from_date = datetime.date(2023, 1, 1)
to_date = datetime.date(2023, 12, 31)
result = reader.read(data_reader.DateRangeCriteria(from_date, to_date))
result.head(3)

Output

	Identifier	Index Date	Open	High	Low	Close	Points Change	Change(%)	Volume	Turnover	P/E	P/B	Div Yield	Date
0	Nifty 50	02-01-2023	18131.7	18215.2	18086.5	18197.5	92.15	0.51	2.56074e+08	1.17931e+11	21.9	4.27	1.26	2023-01-02 00:00:00
1	Nifty Next 50	02-01-2023	42321.1	42409.8	42117.7	42248.4	60.75	0.14	1.34929e+08	4.11122e+10	25.67	4.79	1.76	2023-01-02 00:00:00
2	Nifty 100	02-01-2023	18290.2	18350.9	18237.5	18334.3	75.55	0.41	3.91003e+08	1.59043e+11	22.75	4.41	1.3	2023-01-02 00:00:00

Get daily, monthly and annually aggregrated data

In this example we will use HistoricalDataProcessor class to get data between a date range. HistoricalDataProcessor will also do monthly and annual aggregation of data.

# import classes & setup
# import classes & setup
from markets_insights.dataprocess.data_processor import HistoricalDataProcessor
from markets_insights.datareader.data_reader import NseIndicesReader, DateRangeCriteria
histDataProcessor = HistoricalDataProcessor()

# Fetch and process the data
year_start = datetime.date(2023, 12, 1)
year_end = datetime.date(2023, 12, 31)
result = histDataProcessor.process(NseIndicesReader(), DateRangeCriteria(year_start, year_end))

Displaying the output

We will call get_monthly_data() method to get monthly aggregated data. We can call get_daily_data() for Daily and get_annual_data() for Annual data.

from markets_insights.core.column_definition import BaseColumns

result.get_monthly_data().sort_values(BaseColumns.Date).head(3)

Output

	Identifier	Date	Month	Volume	Turnover	Close	High	Low	Open
0	INDIA VIX	2023-01-31 00:00:00	2023-01	0	0	16.88	19.39	11.6425	14.8675
1012	NIFTY100 ENHANCED ESG	2023-01-31 00:00:00	2023-01	8.12952e+09	4.99243e+12	3352.6	3490.75	3352.6	3472.29
228	NIFTY ALPHA QUALITY VALUE LOW-VOLATILITY 30	2023-01-31 00:00:00	2023-01	2.32482e+09	1.3303e+12	11286.1	11599.4	11214.5	11456.8

Calculating RSI using CalculationPipeline

Below example demonstrates calculating RSI using the calculation pipeline.

# import classes & setup options
import datetime
from markets_insights.datareader.data_reader import BhavCopyReader, DateRangeCriteria
from markets_insights.dataprocess.data_processor import HistoricalDataProcessor, MultiDataCalculationPipelines, CalculationPipelineBuilder, HistoricalDataProcessOptions
from markets_insights.calculations.base import DatePartsCalculationWorker

reader = BhavCopyReader()
options = HistoricalDataProcessOptions()
options.include_monthly_data = False
options.include_annual_data = False
histDataProcessor = HistoricalDataProcessor(options)

# Fetch the data
year_start = datetime.date(2023, 1, 1)
to_date = datetime.date(2023, 12, 31)
result = histDataProcessor.process(reader, DateRangeCriteria(year_start, to_date))

# Prepare calculation pipeline
pipelines = MultiDataCalculationPipelines()
pipelines.set_item('rsi', CalculationPipelineBuilder.create_rsi_calculation_pipeline())
histDataProcessor.set_calculation_pipelines(pipelines)

# Run the pipeline
histDataProcessor.run_calculation_pipelines()

Displaying the output

For displaying the relevant columns in the output we will use column name constants from BaseColumns & CalculatedColumns classes.

from markets_insights.core.column_definition import BaseColumns, CalculatedColumns

result.get_daily_data().sort_values(BaseColumns.Date).tail(3)[[BaseColumns.Identifier, BaseColumns.Date, BaseColumns.Close, CalculatedColumns.RelativeStrengthIndex]]

Output

	Identifier	Date	Close	Rsi
336063	RKFORGE	2023-12-29 00:00:00	725.7	46.9257
329710	RBL	2023-12-29 00:00:00	852.95	54.8479
446931	ZYDUSWELL	2023-12-29 00:00:00	1681.1	72.0492

A real use case: Understand the affect of RSI on price

In this use case, we understand the affect of RSI on the price of equity/stock.

Preparing the data

We perform below steps to prepare our analysis data

• Calculate RSI for each day for all the stocks.
• Add a flag for whenever the RSI crosses the control limits (eg: above 75 and below 30)
• Calculate the highest and lowest price change in the next 1, 3, 5, 7 & 10 trading sessions.
• Find the median for highest price change and lowest price change whenever the RSI crosses the control limits.

# import classes
import datetime
from markets_insights.datareader import data_reader
from markets_insights.dataprocess import data_processor

# Fetch the data
reader = data_reader.BhavCopyReader()
options = data_processor.HistoricalDataProcessOptions(include_monthly_data = False, include_annual_data=False)
histDataProcessor = data_processor.HistoricalDataProcessor(options)

from_date = datetime.date(2023, 12, 1)
to_date = datetime.date(2023, 12, 31)
result = histDataProcessor.process(data_reader.BhavCopyReader(), data_reader.DateRangeCriteria(from_date, to_date))

# prepare calculation pipeline
periods = [1, 7, 15, 30, 45]

pipelines = data_processor.MultiDataCalculationPipelines()
pipelines.set_item('rsi', data_processor.CalculationPipelineBuilder.create_rsi_calculation_pipeline(crossing_above_flag_value = 75, crossing_below_flag_value = 30, window = 14))
pipelines.set_item('foward_looking_fall', data_processor.CalculationPipelineBuilder.create_forward_looking_price_fall_pipeline(periods))
pipelines.set_item('foward_looking_rise', data_processor.CalculationPipelineBuilder.create_forward_looking_price_rise_pipeline(periods))
histDataProcessor.set_calculation_pipelines(pipelines=pipelines)

# run the pipeline and show results
histDataProcessor.run_calculation_pipelines()

daily_data = result.get_daily_data()

# Import constants so its easier to refer to column names
from markets_insights.core.column_definition import BaseColumns, CalculatedColumns

# get names of fwd looking price column names. Since, these column names are auto-generated there no constants for them
fwd_looking_price_fall_cols, fwd_looking_price_rise_cols = [x for x in daily_data.columns if 'Trough' in x], \
    [x for x in daily_data.columns if 'Peak' in x]

Show the median price change % for highest price fall whenever the RSI crosses above

daily_data[
(daily_data[CalculatedColumns.RsiCrossedAbove])
][fwd_looking_price_fall_cols].median()

Output

TroughPercInNext1Sessions     1.417302
TroughPercInNext7Sessions     3.770465
TroughPercInNext15Sessions    4.783065
TroughPercInNext30Sessions    6.070147
TroughPercInNext45Sessions    6.783886
dtype: float64

Show the median price change % for highest price rise whenever the RSI crosses below

daily_data[
(daily_data[CalculatedColumns.RsiCrossedBelow])
][fwd_looking_price_rise_cols].median()

Output

PeakPercInNext1Sessions      3.876428
PeakPercInNext7Sessions      7.600800
PeakPercInNext15Sessions     9.780681
PeakPercInNext30Sessions    13.269178
PeakPercInNext45Sessions    16.275638
dtype: float64

Performing Arithmetic on Readers

from markets_insights.datareader import data_reader
from markets_insights.core.core import IdentifierFilter

date_criteria = data_reader.DateRangeCriteria(datetime.date(2023, 1, 1), datetime.date(2023, 12, 31))

indices_reader = data_reader.NseIndicesReader()
vix_reader = data_reader.NseIndicesReader().set_filter(IdentifierFilter("India VIX"))
op_reader = indices_reader / vix_reader
data = op_reader.read(date_criteria).query(str(IdentifierFilter("Nifty 50 / India VIX")))

Output

Extending the Framework: Creating a DataReader

In this example we will create a new data reader to read data for Nasdaq listed equities. We will use yfinance python library for this.

Import classes

from markets_insights.datareader.data_reader import DateRangeDataReader
from markets_insights.core.core import Instrumentation
from markets_insights.core.column_definition import BaseColumns

import yfinance as yf
import pandas

Create reader class

We will create a class that extends the base reader. yfinance library can read data for a range. So, we will extend DateRangeDataReader class. With yfinance library, we have to specify which equity/tickers we want to download. For the sake of this example, we will download for top 7 companies of Nasdaq.

class NasdaqDataReader (DateRangeSourceDataReader):
  def __init__(self, tickers: list = ['AAPL', 'GOOGL', 'MSFT', 'AMZN', 'META', 'TSLA', 'NVDA']):
    super().__init__(reader=None)
    self.tickers = tickers
    self.name = "NasdaqDataReader"

  @Instrumentation.trace(name="NasdaqDataReader.read")
  def read(self, criteria: DateRangeCriteria):
    df_list = list()
    for ticker in self.tickers:
        data = yf.download(ticker, group_by="Ticker", start=criteria.from_date, end=criteria.to_date)
        data['ticker'] = ticker
        df_list.append(data)

    # combine all dataframes into a single dataframe
    df = pandas.concat(df_list)

    final_data = df.reset_index().rename(columns = self.get_column_name_mappings())
    final_data[BaseColumns.Date] = pandas.to_datetime(final_data[BaseColumns.Date])
    return final_data
  
  def get_column_name_mappings(self):
    return {
      'ticker': BaseColumns.Identifier,
      'OPEN': BaseColumns.Open,
      'HIGH': BaseColumns.High,
      'LOW': BaseColumns.Low,
      'CLOSE': BaseColumns.Close
    }

Notice here we are renaming the columns to standard column names so that the calculation pipeline can read them properly.

Running the calculation pipeline

The calculation pipeline will not be different except we will pass NasdaqDataReader instance.

# import classes & setup options
import datetime
from markets_insights.dataprocess.data_processor import HistoricalDataProcessor, MultiDataCalculationPipelines, CalculationPipelineBuilder, HistoricalDataProcessOptions, DateRangeCriteria
from markets_insights.calculations.base import DatePartsCalculationWorker

reader = NasdaqDataReader()
options = HistoricalDataProcessOptions()
options.include_monthly_data = False
options.include_annual_data = False
histDataProcessor = HistoricalDataProcessor(options)

# Fetch the data
year_start = datetime.date(2023, 1, 1)
to_date = datetime.date(2023, 12, 31)
result = histDataProcessor.process(reader, DateRangeCriteria(year_start, to_date))

# Prepare calculation pipeline
pipelines = MultiDataCalculationPipelines()
pipelines.set_item('date_parts', CalculationPipelineBuilder.create_pipeline_for_worker(DatePartsCalculationWorker()))
pipelines.set_item('rsi', CalculationPipelineBuilder.create_rsi_calculation_pipeline())
histDataProcessor.set_calculation_pipelines(pipelines)

# Run the pipeline
histDataProcessor.run_calculation_pipelines()

Here's the code to display results

from markets_insights.core.column_definition import CalculatedColumns

result.get_daily_data() \
  .sort_values(
    [BaseColumns.Date, BaseColumns.Identifier]
  )[
    [BaseColumns.Identifier, BaseColumns.Date, BaseColumns.Close, 
     CalculatedColumns.RelativeStrengthIndex]
  ] \
  .tail(5)

Output

	Identifier	Date	Close	Rsi
248	AAPL	2023-12-28 00:00:00	193.58	54.4815
497	AMZN	2023-12-28 00:00:00	153.38	63.9387
746	GOOGL	2023-12-28 00:00:00	140.23	61.585
995	META	2023-12-28 00:00:00	358.32	70.2377
1244	MSFT	2023-12-28 00:00:00	375.28	56.909
1493	NVDA	2023-12-28 00:00:00	495.22	58.305
1742	TSLA	2023-12-28 00:00:00	253.18	55.9788

Extending the Framework: Creating a CalculationWorker

In this example, we will create a CalculationWorker to calcualte the Fibonacci Retracement level for any equity or index. Finbonacci Retracement levels are based on a time window and a level (26.3%, 50% etc). So, these will become input to our CalculationWorker. Lets call this worker as FibnocciRetracementCalculationWorker

Implement the worker class. The important aspect here is to override the add_calculated_columns() method

## import modules
from markets_insights.calculations.base import CalculationWorker
from markets_insights.core.core import Instrumentation
from markets_insights.calculations.base import BaseColumns
import pandas

class FibonacciRetracementCalculationWorker (CalculationWorker):
  def __init__(self, time_window: int, level_perct: float):
    self._time_window = time_window
    self._level = level_perct / 100
    self._column_name = 'Fbr' + str(level_perct)

  @Instrumentation.trace(name="FibnocciRetracementCalculationWorker")
  def add_calculated_columns(self, data: pandas.DataFrame):
    identifier_grouped_data: pandas.DataFrame = data.groupby(BaseColumns.Identifier)
    #Since, our dataframe may contain data for multiple symbols, we need to first group them by Identifier
    data[self._column_name] = identifier_grouped_data[BaseColumns.Close].transform(
        lambda x: 
          x.rolling(self._time_window).max() - 
          (
            (x.rolling(self._time_window).max() - x.rolling(self._time_window).min())  * self._level
          )
      )

Create pipline with the FibnocciRetracementCalculationWorker and run

Now, that our worker is created let us use it in a calculation pipeline. We can use it with any instrument (index, stock) that are supported. Event for the Nasdaq instruments that were supported in earlier examples. For this example, let us take NSE Indexes.

# Create pipline with the FibnocciRetracementCalculationWorker and run 
from markets_insights.datareader.data_reader import NseIndicesReader, DateRangeCriteria
from markets_insights.dataprocess.data_processor import HistoricalDataProcessor, HistoricalDataProcessOptions, \
  MultiDataCalculationPipelines, CalculationPipeline
histDataProcessor = HistoricalDataProcessor(HistoricalDataProcessOptions(include_monthly_data=False, include_annual_data=False))

# Fetch the data
result = histDataProcessor.process(NseIndicesReader(), DateRangeCriteria(datetime.date(2023, 12, 1), datetime.date(2023, 12, 31)))

# Prepare calculation pipeline
fbr50_worker = FibonacciRetracementCalculationWorker(time_window=7, level_perct=50)
pipelines = MultiDataCalculationPipelines()
histDataProcessor.set_calculation_pipelines(
  CalculationPipeline(
    workers = [fbr50_worker]
  )
)

# Run the pipeline and get data
histDataProcessor.run_calculation_pipelines()

Display the results.

Since our time window was 15 days. So, the calculation result for first 14 days will not be available. We will look at the last 10 records with tail(5)

result.get_daily_data()[[
  BaseColumns.Identifier, BaseColumns.Date, BaseColumns.Close, fbr50_worker._column_name
]].tail(5)

Output

	Identifier	Date	Close	Fbr50
2141	NIFTY COMPOSITE G-SEC INDEX	2023-12-29 00:00:00	2602.3	2599.48
2142	NIFTY 10 YR BENCHMARK G-SEC	2023-12-29 00:00:00	2232.79	2230.24
2143	NIFTY MIDCAP SELECT	2023-12-29 00:00:00	10397.5	10209.1
2144	NIFTY ALPHA LOW-VOLATILITY 30	2023-12-29 00:00:00	23373.2	22886
2145	NIFTY50 USD	2023-12-29 00:00:00	9048.9	8941.77