Calculating Higher Highs & Lower Lows in Stock Market Dataframe

I have a DataFrame containing intraday, 5 min data, for 1 month, 23 days trading days. Around 1725 rows, so 75 rows per day. There is a signal column, that gives buy/sell signal based on ST for now, but I might use another indicator later on.

PHASE 2 (in the Code added below) is where the code has to be corrected/implemented. I started working on it, but been stuck due to my lack of pandas knowledge.

1. So I wanna create a new column, that only captures "High" column values, if the "High" value at any point of that same day is greater than the "High" value corresponding to the "buy" signal row.

2. And also, create a new column, that only captures "Low" column values, if the "Low" value at any point of that same day is Lower than the "Low" value corresponding to "buy" signal row. Condition 2 is only checked if condition 1 is satisfied.

These 2 columns may help in analyzing the data, or for calculation. Column creation is not important but getting the necessary values for performing the calculations is necessary.

3. Need a column that calculates returns.

   3a. If Condition 1 didn't happen, return 0.

   3b. If Condition 1 happened, but condition 2 didn't happen, return (last "Close" of day - "High" of "buy" signal row)

   3c. If Condition 1 happened, AND condition 2 happened, return ("Low" of "buy" row - "High" of "buy" row)

Adding some example scenarios below -

Here the Higher High and Lower Low both happened in same row.

return = Condition 3c, i.e. ["Low" of "buy" row - "High" of "buy" row]

Another example below -

Here the Higher High happened first, then after a few rows, Lower Low happened.

return = Condition 3c, i.e. ["Low" of "buy" row - "High" of "buy" row]

In cases where Lower Low doesn't happen, we take the last "Close" of that day.

Return = Condition 3b, i.e. ["Close" at 3 pm of that day - "High" of "buy" signal row]

All of this above is for "buy" signal condition. I'll have to do it for "sell" as well. But if I can learn how to do it for one, I'll be able to do the other.

Code & Sample Data -

# -*- coding: utf-8 -*-
"""
Created on Mon Jun 15 


"""

import datetime as dt
import yfinance as yf
import pandas as pd
import numpy as np
import pandas_ta as ta
import copy


tickers = ["LT.NS","BRITANNIA.NS","CIPLA.NS","HINDALCO.NS"]

ohlc_intraday = {}

today = dt.datetime.today()
end = today - dt.timedelta(1)
start = today - dt.timedelta(20)

# looping over tickers and creating a dataframe with ohlcv, intraday tf
for ticker in tickers:
    temp = yf.download(ticker, period="1mo",interval = "5m")
    temp.dropna(how="any",inplace=True)
    # temp = temp[temp["Volume"] > 0]
    ohlc_intraday[ticker] = temp
    # ohlc_intraday[ticker].to_csv("supertrend_data.csv")

    


tickers = ohlc_intraday.keys()
# dont run above code again & again
# run below code    


# Phase 1 - Data Collection & Signal Generation
ohlc_dict = copy.deepcopy(ohlc_intraday)
tickers_signal = {}
tickers_ret = {}


for ticker in tickers:
    
    ohlc_dict[ticker]["datetime"] = pd.to_datetime(ohlc_dict[ticker].index)
    ohlc_dict[ticker]["hour"] = ohlc_dict[ticker]["datetime"].dt.hour
    ohlc_dict[ticker]["date"] = ohlc_dict[ticker]["datetime"].dt.date
    unique_days = len(ohlc_dict[ticker]["date"].unique())
    num_5min_candles = 75 # number of candles in intraday
    ohlc_dict[ticker]["candle"] = np.array(range(len(ohlc_dict[ticker])))%(num_5min_candles) + 1
    
    
    # Supertrend indicator's value column
    ohlc_dict[ticker]["ST"] = ta.supertrend(high=ohlc_dict[ticker]["High"],low=ohlc_dict[ticker]["Low"],
                                  close=ohlc_dict[ticker]["Close"],length=7,
                                  multiplier=3)["SUPERT_7_3.0"]
    

    def st_levels(close,stlevel):
        if close > stlevel:
            return "buy"
        elif close < stlevel:
            return "sell"
        
    
    # capture Supertrend buy/sell signals    
    ohlc_dict[ticker]["signal"] = np.vectorize(st_levels)(ohlc_dict[ticker]["Close"],ohlc_dict[ticker]["ST"])
    
    # keep only first occurrence of signal, remove consecutive duplicated signals
    ohlc_dict[ticker].loc[ohlc_dict[ticker]["signal"].eq(ohlc_dict[ticker]["signal"].shift()), "signal"] = "no"
    
    # supertrend length; first 7 rows will contain "no" since ST period length is 7
    ohlc_dict[ticker]["signal"][:7] = "no"
    
    
    # capturing returns
    tickers_signal[ticker] = ""
    tickers_ret[ticker] = []
    # ohlc_dict[ticker].to_csv("supertrend_data.csv")


# Phase 2 - Calculating returns for intraday -
for ticker in tickers:
    print("Calculating returns for ", ticker)
    # ticker = "CIPLA.NS"
    for i in range(len(ohlc_dict[ticker])):
        if tickers_signal[ticker] == "":
            tickers_ret[ticker].append(0)
            # check if signal candle High/Low is broken for Buy/Sell Scenario respectively
            if ohlc_dict[ticker]["High"][i] >= ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["High"].values[0]:
                tickers_signal[ticker] = "BUY"
            elif ohlc_dict[ticker]["Low"][i] <= ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="sell"]["Low"].values[0]:    
                tickers_signal[ticker] = "SELL"
        
        elif tickers_signal[ticker] == "BUY":
            
            # calculating returns for StopLoss            
            if ohlc_dict[ticker]["Low"][i] < ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["Low"].values[0]: # and add that both candles be on same day
                tickers_signal[ticker] = ""
                tickers_ret[ticker].append(((ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["Low"].values[0] - ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["High"].values[0])/(ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["High"].values[0])) - 1)
            # calculating returns for 70th candle close in intraday
            else:
                tickers_ret[ticker].append(((ohlc_dict[ticker][ohlc_dict[ticker]['candle']==70]["Close"].values[0] - ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["High"].values[0])/(ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["High"].values[0])) - 1)
                                
        elif tickers_signal[ticker] == "SELL":
            # calculating returns for StopLoss
            if ohlc_dict[ticker]["High"][i] > ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["High"].values[0]:
                tickers_signal[ticker] = ""
                tickers_ret[ticker].append(((ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["Low"].values[0] - ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["High"].values[0])/(ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["Low"].values[0])) - 1)
            
            # calculating returns for 70th candle close in intraday
            else:
                tickers_ret[ticker].append(((ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["Low"].values[0] - ohlc_dict[ticker][ohlc_dict[ticker]['candle']==70]["Close"].values[0])/(ohlc_dict[ticker][ohlc_dict[ticker]['signal']=="buy"]["High"].values[0])) - 1)


    ohlc_dict[ticker]["ret"] = np.array(tickers_ret[ticker])
                
                

Below data is for HINDALCO.NS stock -

UPDATE - In case of "buy" signal happens, and Trade is live, waiting to exit in loss or profit, and if opposite signal "sell" appears, exit trade at that "sell" signal "Close" value.

If this is too complicated, just ignore the signal change happening on same day.

PHASE 2 is where the code has to be corrected/implemented. Checking if signal was generated, if Higher Highs / Lower Lows happened, or just exiting at Market Close (3 pm) and calculate returns accordingly.

Currently as per my code, I'm trying to take "Close" of 70th candle, which is around 15:05 pm. But please feel free to modify/improve it so the condition takes 15:00 pm Close value, either using candle count, or using datetime value.

I know this is a huge favor I'm asking, but I have been trying to get this done for over 2 weeks and couldn't find any help. Lot of tutorials and solutions are on basic, standardized methods. This is a bit unconventional, and so unable to solve it on my own.