siglab-py 0.1.29__py3-none-any.whl → 0.6.12__py3-none-any.whl

siglab_py/util/market_data_util.py

@@ -1,13 +1,19 @@
+import logging
+import incremental
 import tzlocal
 from datetime import datetime, timezone
+import time
 from typing import List, Dict, Union, NoReturn, Any, Tuple
 from pathlib import Path
 import math
 import pandas as pd
 import numpy as np
+import asyncio
+from tabulate import tabulate
 
 from ccxt.base.exchange import Exchange as CcxtExchange
-from ccxt import deribit
+import ccxt
+import ccxt.pro as ccxtpro
 
 # https://www.analyticsvidhya.com/blog/2021/06/download-financial-dataset-using-yahoo-finance-in-python-a-complete-guide/
 from yahoofinancials import YahooFinancials
@@ -15,72 +21,271 @@ from yahoofinancials import YahooFinancials
 # yfinance allows intervals '1m', '5m', '15m', '1h', '1d', '1wk', '1mo'. yahoofinancials not as flexible
 import yfinance as yf
 
+from siglab_py.util.retry_util import retry
 from siglab_py.exchanges.futubull import Futubull
+from siglab_py.exchanges.any_exchange import AnyExchange
+from siglab_py.exchanges.deribit import Deribit, DeribitAsync
+from siglab_py.exchanges.binance import Binance, BinanceAsync
+
+def instantiate_exchange(
+    exchange_name : str,
+    api_key : Union[str, None] = None,
+    secret : Union[str, None]  = None,
+    passphrase : Union[str, None] = None,
+    default_type : Union[str, None] = 'spot',
+    default_sub_type : Union[str, None] = None,
+    rate_limit_ms : float = 100
+) -> Union[AnyExchange, None]:
+    exchange_name = exchange_name.lower().strip()
+
+    # Look at ccxt exchange.describe. under 'options' \ 'defaultType' (and 'defaultSubType') for what markets the exchange support.
+    # https://docs.ccxt.com/en/latest/manual.html#instantiation
+    exchange_params : Dict[str, Any]= {
+                        'apiKey' : api_key,
+                        'secret' : secret,
+                        'enableRateLimit'  : True,
+                        'rateLimit' : rate_limit_ms,
+                        'options' : {
+                            'defaultType' : default_type
+                        }
+                    }
+    
+    if default_sub_type:
+        exchange_params['defaultSubType'] = default_sub_type
+
+    if api_key:
+        exchange_params['apiKey'] = api_key
+    if secret:
+        exchange_params['secret'] = secret
+    if passphrase:
+        exchange_params['passphrase'] = passphrase
+
+    if exchange_name=='binance':
+        exchange = Binance(exchange_params)  # type: ignore
+    elif exchange_name=='bybit':
+        exchange = ccxt.bybit(exchange_params) # type: ignore
+    elif exchange_name=='okx':
+        exchange = ccxt.okx(exchange_params) # type: ignore
+    elif exchange_name=='deribit':
+        exchange = Deribit(exchange_params)  # type: ignore
+    elif exchange_name=='hyperliquid':
+        exchange = ccxt.hyperliquid(
+            {
+                "walletAddress" : api_key, # type: ignore
+                "privateKey" : secret, 
+                'enableRateLimit'  : True,
+                'rateLimit' : rate_limit_ms
+            }
+        )  
+    else:
+        raise ValueError(f"Unsupported exchange {exchange_name}.")
+
+    exchange.load_markets() # type: ignore
+
+    return exchange # type: ignore
+
+async def async_instantiate_exchange(
+    gateway_id : str,
+    api_key : str,
+    secret : str,
+    passphrase : str,
+    default_type : Union[str, None] = 'spot',
+    default_sub_type : Union[str, None] = None,
+    rate_limit_ms : float = 100,
+    verbose : bool = False
+) -> Union[AnyExchange, None]:
+    exchange : Union[AnyExchange, None] = None
+    exchange_name : str = gateway_id.split('_')[0]
+    exchange_name =exchange_name.lower().strip()
+
+    # Look at ccxt exchange.describe. under 'options' \ 'defaultType' (and 'defaultSubType') for what markets the exchange support.
+    # https://docs.ccxt.com/en/latest/manual.html#instantiation
+    exchange_params : Dict[str, Any]= {
+                        'apiKey' : api_key,
+                        'secret' : secret,
+                        'enableRateLimit'  : True,
+                        'rateLimit' : rate_limit_ms,
+                        'options' : {
+                            'defaultType' : default_type
+                        },
+                        'verbose': verbose
+                    }
+    
+    if default_sub_type:
+        exchange_params['defaultSubType'] = default_sub_type
+
+    if exchange_name=='binance':
+        # spot, future, margin, delivery, option
+        # https://github.com/ccxt/ccxt/blob/master/python/ccxt/binance.py#L1298
+        exchange = BinanceAsync(exchange_params)  # type: ignore
+    elif exchange_name=='bybit':
+        # spot, linear, inverse, futures
+        # https://github.com/ccxt/ccxt/blob/master/python/ccxt/bybit.py#L1041
+        exchange = ccxtpro.bybit(exchange_params) # type: ignore
+    elif exchange_name=='okx':
+        # 'funding', spot, margin, future, swap, option
+        # https://github.com/ccxt/ccxt/blob/master/python/ccxt/okx.py#L1144
+        exchange_params['password'] = passphrase
+        exchange = ccxtpro.okx(exchange_params) # type: ignore
+    elif exchange_name=='deribit':
+        # spot, swap, future
+        # https://github.com/ccxt/ccxt/blob/master/python/ccxt/deribit.py#L360
+        exchange = DeribitAsync(exchange_params)  # type: ignore
+    elif exchange_name=='kraken':
+        exchange = ccxtpro.kraken(exchange_params) # type: ignore
+    elif exchange_name=='hyperliquid':
+        '''
+        https://app.hyperliquid.xyz/API
+        
+        defaultType from ccxt: swap
+            https://github.com/ccxt/ccxt/blob/master/python/ccxt/hyperliquid.py#L225
+        
+        How to integrate? You can skip first 6 min: https://www.youtube.com/watch?v=UuBr331wxr4&t=363s
+
+        Example, 
+            API credentials created under "\ More \ API":
+                    Ledger Arbitrum Wallet Address: 0xAAAAA <-- This is your Ledger Arbitrum wallet address with which you connect to Hyperliquid. 
+                    API Wallet Address 0xBBBBB <-- Generated
+                    privateKey 0xCCCCC
+
+        Basic connection via CCXT:
+            import asyncio
+            import ccxt.pro as ccxtpro
+
+            async def main():
+                rate_limit_ms = 100
+                exchange_params = {
+                    "walletAddress" : "0xAAAAA", # Ledger Arbitrum Wallet Address here! Not the generated address.
+                    "privateKey" : "0xCCCCC"
+                }
+                exchange = ccxtpro.hyperliquid(exchange_params) 
+                balances = await exchange.fetch_balance()
+                print(balances)
+
+            asyncio.run(main())
+        '''
+        exchange = ccxtpro.hyperliquid(
+            {
+                "walletAddress" : api_key,
+                "privateKey" : secret,
+                'enableRateLimit'  : True,
+                'rateLimit' : rate_limit_ms,
+                'verbose': verbose
+            }  # type: ignore
+        )
+    else:
+        raise ValueError(f"Unsupported exchange {exchange_name}, check gateway_id {gateway_id}.")
+
+    await exchange.load_markets() # type: ignore
+
+    '''
+    Is this necessary? The added trouble is for example bybit.authenticate requires arg 'url'. binance doesn't. And fetch_balance already test credentials.
+
+    try:
+        await exchange.authenticate() # type: ignore
+    except Exception as swallow_this_error:
+        pass
+    '''
+
+    return exchange
 
 def timestamp_to_datetime_cols(pd_candles : pd.DataFrame):
-    pd_candles['datetime'] = pd_candles['timestamp_ms'].apply(
-        lambda x: datetime.fromtimestamp(int(x.timestamp()) if isinstance(x, pd.Timestamp) else int(x / 1000))
-    )
+    def _fix_timestamp_ms(x):
+        if isinstance(x, pd.Timestamp):
+            return int(x.value // 10**6)
+        elif isinstance(x, np.datetime64):
+            return int(x.astype('int64') // 10**6)
+        elif isinstance(x, (int, float)):
+            x = int(x)
+            if len(str(abs(x))) == 13:
+                return x
+            else:
+                return int(x * 1000)
+        else:
+            raise ValueError(f"Unsupported type {type(x)} for timestamp conversion")
+    pd_candles['timestamp_ms'] = pd_candles['timestamp_ms'].apply(_fix_timestamp_ms)
+    pd_candles['datetime'] = pd_candles['timestamp_ms'].apply(lambda x: datetime.fromtimestamp(int(x/1000)))
     pd_candles['datetime'] = pd.to_datetime(pd_candles['datetime'])
-    pd_candles['datetime'] = pd_candles['datetime'].dt.tz_localize(None)
+    pd_candles['datetime'] = pd_candles['datetime'].dt.tz_localize(None)  # type: ignore
     pd_candles['datetime_utc'] = pd_candles['timestamp_ms'].apply(
         lambda x: datetime.fromtimestamp(int(x.timestamp()) if isinstance(x, pd.Timestamp) else int(x / 1000), tz=timezone.utc)
     )
     
     # This is to make it easy to do grouping with Excel pivot table
-    pd_candles['year'] = pd_candles['datetime'].dt.year
-    pd_candles['month'] = pd_candles['datetime'].dt.month
-    pd_candles['day'] = pd_candles['datetime'].dt.day
-    pd_candles['hour'] = pd_candles['datetime'].dt.hour
-    pd_candles['minute'] = pd_candles['datetime'].dt.minute
-    pd_candles['dayofweek'] = pd_candles['datetime'].dt.dayofweek  # dayofweek: Monday is 0 and Sunday is 6
+    pd_candles['year'] = pd_candles['datetime'].dt.year  # type: ignore
+    pd_candles['month'] = pd_candles['datetime'].dt.month  # type: ignore
+    pd_candles['day'] = pd_candles['datetime'].dt.day  # type: ignore
+    pd_candles['hour'] = pd_candles['datetime'].dt.hour  # type: ignore
+    pd_candles['minute'] = pd_candles['datetime'].dt.minute  # type: ignore
+    pd_candles['dayofweek'] = pd_candles['datetime'].dt.dayofweek  # type: ignore dayofweek: Monday is 0 and Sunday is 6
 
     pd_candles['week_of_month'] = pd_candles['timestamp_ms'].apply(
-        lambda x: timestamp_to_week_of_month(x)
+        lambda x: timestamp_to_week_of_month(int(x/1000))
     )
 
     pd_candles['apac_trading_hr'] = pd_candles['timestamp_ms'].apply(
-        lambda x: "APAC" in timestamp_to_active_trading_regions(x)
+        lambda x: "APAC" in timestamp_to_active_trading_regions(int(x/1000))
     )
     pd_candles['emea_trading_hr'] = pd_candles['timestamp_ms'].apply(
-        lambda x: "EMEA" in timestamp_to_active_trading_regions(x)
+        lambda x: "EMEA" in timestamp_to_active_trading_regions(int(x/1000))
     )
     pd_candles['amer_trading_hr'] = pd_candles['timestamp_ms'].apply(
-        lambda x: "AMER" in timestamp_to_active_trading_regions(x)
+        lambda x: "AMER" in timestamp_to_active_trading_regions(int(x/1000))
     )
 
+    pd_candles['timestamp_ms_gap'] = pd_candles['timestamp_ms'] - pd_candles['timestamp_ms'].shift(1)
+    
+    # Depending on asset, minutes bar may have gaps
+    timestamp_ms_gap_median = pd_candles['timestamp_ms_gap'].median()
+    NUM_MS_IN_1HR = 60*60*1000
+    if timestamp_ms_gap_median>=NUM_MS_IN_1HR:
+        num_rows_with_expected_gap = pd_candles[~pd_candles.timestamp_ms_gap.isna()][pd_candles.timestamp_ms_gap==timestamp_ms_gap_median].shape[0]
+        assert(num_rows_with_expected_gap/pd_candles.shape[0]>0.9)
+    pd_candles.drop(columns=['timestamp_ms_gap'], inplace=True)
+
+'''
+APAC (Asia-Pacific) Trading Hours
+    UTC 21:00 - 09:00 (approximate range)
+    Major financial centers: Tokyo, Hong Kong, Singapore, Sydney
+
+EMEA (Europe, Middle East, Africa) Trading Hours
+    UTC 07:00 - 16:00 (approximate range)
+    Major financial centers: London, Frankfurt, Paris, Zurich, Dubai
+
+US Trading Hours
+    UTC 13:00 - 22:00 (approximate range)
+    Major financial centers: New York, Chicago
+    Key markets: NYSE, NASDAQ
+
+utcnow and utcfromtimestamp been deprecated in Python 3.12 
+https://www.pythonmorsels.com/converting-to-utc-time/
+
+Example, UTC 23:00 is 3rd hour in APAC trading session
+    utc_hour = 23
+    i = get_regions_trading_utc_hours()['APAC'].index(utc_hour)
+    assert(i==2)
+'''
+def get_regions_trading_utc_hours():
+    return {
+        'APAC' : [21,22,23,0,1,2,3,4,5,6,7,8,9],
+        'EMEA' : [7,8,9,10,11,12,13,14,15,16],
+        'AMER' : [13,14,15,16,17,18,19,20,21,22]
+    }
+			
 def timestamp_to_active_trading_regions(
         timestamp_ms : int
 ) -> List[str]:
-    
-    '''
-    APAC (Asia-Pacific) Trading Hours
-        UTC 22:00 - 09:00 (approximate range)
-        Major financial centers: Tokyo, Hong Kong, Singapore, Sydney
-
-    EMEA (Europe, Middle East, Africa) Trading Hours
-        UTC 07:00 - 16:00 (approximate range)
-        Major financial centers: London, Frankfurt, Paris, Zurich, Dubai
-
-    US Trading Hours
-        UTC 13:30 - 20:00 (approximate range)
-        Major financial centers: New York, Chicago
-        Key markets: NYSE, NASDAQ
-
-    utcnow and utcfromtimestamp been deprecated in Python 3.12 
-    https://www.pythonmorsels.com/converting-to-utc-time/
-    '''
     active_trading_regions : List[str] = []
 
     dt_utc = datetime.fromtimestamp(int(timestamp_ms / 1000), tz=timezone.utc)
     utc_hour = dt_utc.hour
-    if (utc_hour >= 22) or (utc_hour <= 9):
+    if utc_hour in get_regions_trading_utc_hours()['APAC']:
         active_trading_regions.append("APAC") 
 
-    if 7 <= utc_hour <= 16:
+    if utc_hour in get_regions_trading_utc_hours()['EMEA']:
         active_trading_regions.append("EMEA")
 
-    if 13 <= utc_hour <= 20:
+    if utc_hour in get_regions_trading_utc_hours()['AMER']:
         active_trading_regions.append("AMER")
 
     return active_trading_regions
@@ -113,6 +318,17 @@ def fix_column_types(pd_candles : pd.DataFrame):
     pd_candles.reset_index(drop=True, inplace=True)
     pd_candles.sort_values("datetime", inplace=True)
 
+def interval_to_ms(interval : str) -> int:
+    interval_ms : int = 0
+    if interval=="d":
+        interval_ms = 24*60*60*1000
+    elif interval=="h":
+        interval_ms = 60*60*1000
+    elif interval=="m":
+        interval_ms = 60*1000
+
+    return interval_ms
+
 '''
 https://polygon.io/docs/stocks
 '''
@@ -168,7 +384,7 @@ class NASDAQExchange:
             pd_daily_candles['low'] = pd_daily_candles['low'].astype(str).str.replace('$','')
             pd_daily_candles['close'] = pd_daily_candles['close'].astype(str).str.replace('$','')
             pd_daily_candles['datetime']= pd.to_datetime(pd_daily_candles['datetime'])
-            pd_daily_candles['timestamp_ms'] = pd_daily_candles.datetime.values.astype(np.int64) // 10 ** 6
+            pd_daily_candles['timestamp_ms'] = pd_daily_candles.datetime.values.astype(np.int64) // 10 ** 6  # type: ignore
             pd_daily_candles['symbol'] = symbol
             pd_daily_candles['exchange'] = 'nasdaq'
             fix_column_types(pd_daily_candles)
@@ -187,7 +403,7 @@ class NASDAQExchange:
                 )
 
                 # When you fill foward, a few candles before start date can have null values (open, high, low, close, volume ...)
-                first_candle_dt = pd_hourly_candles[(~pd_hourly_candles.close.isna())  & (pd_hourly_candles['datetime'].dt.time == pd.Timestamp('00:00:00').time())].iloc[0]['datetime']
+                first_candle_dt = pd_hourly_candles[(~pd_hourly_candles.close.isna())  & (pd_hourly_candles['datetime'].dt.time == pd.Timestamp('00:00:00').time())].iloc[0]['datetime']  # type: ignore
                 pd_hourly_candles = pd_hourly_candles[pd_hourly_candles.datetime>=first_candle_dt]
                 exchange_candles[symbol] = pd_hourly_candles
 
@@ -251,6 +467,45 @@ class YahooExchange:
 
         return exchange_candles
 
+def aggregate_candles(
+    interval : str,
+    pd_candles : pd.DataFrame
+) -> pd.DataFrame:
+    if interval[-1]=='m':
+        # 'm' for pandas means months!
+        interval = interval.replace('m','min')
+    pd_candles.set_index('datetime', inplace=True)
+    pd_candles_aggregated = pd_candles.resample(interval).agg({
+        'exchange' : 'first',
+        'symbol' : 'first',
+        'timestamp_ms' : 'first',
+        
+        'open': 'first',
+        'high': 'max',
+        'low': 'min',
+        'close': 'last',
+        'volume': 'sum',
+
+        'datetime_utc' : 'first',
+        'year' : 'first',
+        'month' : 'first',
+        'day' : 'first',
+        'hour' : 'first',
+        'minute' : 'first',
+        'dayofweek' : 'first',
+        'week_of_month' : 'first',
+
+        'apac_trading_hr' : 'first',
+        'emea_trading_hr' : 'first',
+        'amer_trading_hr' : 'first',
+
+        'pct_chg_on_close' : 'sum',
+
+    })
+    pd_candles.reset_index(inplace=True)
+    pd_candles_aggregated.reset_index(inplace=True)
+    return pd_candles_aggregated
+    
 def fetch_historical_price(
     exchange,
     normalized_symbol : str,
@@ -297,15 +552,21 @@ def fetch_candles(
     validation_max_gaps : int = 10,
     validation_max_end_date_intervals : int = 1
 ) -> Dict[str, Union[pd.DataFrame, None]]:
+    exchange_candles = { '' : None }
+    num_intervals = int(candle_size.replace(candle_size[-1],''))
+
+    if end_ts>datetime.now().timestamp():
+        end_ts = int(datetime.now().timestamp())
+
     if type(exchange) is YahooExchange:
-        return exchange.fetch_candles(
+        exchange_candles = exchange.fetch_candles(
                             start_ts=start_ts,
                             end_ts=end_ts,
                             symbols=normalized_symbols,
                             candle_size=candle_size
                         )
     elif type(exchange) is NASDAQExchange:
-        return exchange.fetch_candles(
+        exchange_candles = exchange.fetch_candles(
                             start_ts=start_ts,
                             end_ts=end_ts,
                             symbols=normalized_symbols,
@@ -322,79 +583,153 @@ def fetch_candles(
             pd_candles = exchange_candles[symbol]
             if not pd_candles is None:
                 fix_column_types(pd_candles) # You don't want to do this from Futubull as you'd need import Futubull from there: Circular references
-        return exchange_candles
+                
     elif issubclass(exchange.__class__, CcxtExchange):
-        return _fetch_candles_ccxt(
+        exchange_candles = _fetch_candles_ccxt(
             start_ts=start_ts,
             end_ts=end_ts,
             exchange=exchange,
             normalized_symbols=normalized_symbols,
             candle_size=candle_size,
-            logger=logger,
-            num_candles_limit=num_candles_limit,
-            cache_dir=cache_dir,
-            list_ts_field=list_ts_field
+            num_candles_limit=num_candles_limit
         )
-    return { '' : None }
+    if num_intervals!=1:
+        for symbol in exchange_candles:
+            if not exchange_candles[symbol] is None:
+                exchange_candles[symbol] = aggregate_candles(candle_size, exchange_candles[symbol]) #  type: ignore
+
+    # For invalid rows missing timestamps, o/h/l/c/v, fill forward close, set volume to zero.
+    for symbol in exchange_candles:
+        pd_candles = exchange_candles[symbol]
+        
+        if pd_candles is not None:
+            mask_invalid_candles = pd_candles["timestamp_ms"].isna()
+            if mask_invalid_candles.any():
+                pd_invalid_candles = pd_candles[mask_invalid_candles]
+
+                if logger is not None:
+                    logger.warning(f"Dropping {pd_invalid_candles.shape[0]}/{pd_candles.shape[0]} rows from {symbol} candles (null timestamp_ms)") # type: ignore
+                    logger.warning(f"{tabulate(pd_invalid_candles, headers='keys', tablefmt='psql')}") # type: ignore
+                    
+                def _to_timestamp_ms(dt):
+                    if pd.isna(dt):
+                        return pd.NA
+                    if isinstance(dt, str):
+                        dt = pd.to_datetime(dt)
+                    return int(dt.timestamp() * 1000)
+                
+                pd_candles.loc[mask_invalid_candles, "timestamp_ms"] = pd_candles.loc[mask_invalid_candles, "datetime"].apply(_to_timestamp_ms)
+
+                pd_candles["close"] = pd_candles["close"].ffill()
+                pd_candles.loc[mask_invalid_candles, ["open", "high", "low"]] = pd_candles.loc[
+                                                                                    mask_invalid_candles, ["close"]
+                                                                                ]
+                pd_candles.loc[mask_invalid_candles, "volume"] = 0.0
+
+    return exchange_candles # type: ignore
+
+'''
+Find listing date https://gist.github.com/mr-easy/5185b1dcdd5f9f908ff196446f092e9b
+
+Usage:
+    listing_ts = find_start_time(exchange, 'HYPE/USDT:USDT', int(datetime(2024,1,1).timestamp()*1000), int(datetime(2025,5,1).timestamp()*1000), '1h')
 
+Caveats: 
+1) If listing date lies outside [start_time, end_time], this function will stackoverflow, 
+2) Even if not, it's still very time consuming.
+
+Alternative: market['created']
+'''
+def search_listing_ts(exchange, symbol, start_time, end_time, timeframe):
+    mid_time = (start_time + end_time)//2
+    if(mid_time == start_time): return mid_time+1
+    ohlcv = exchange.fetch_ohlcv(symbol, timeframe, mid_time, limit=1)
+    time.sleep(1)
+    if(len(ohlcv) == 0):
+        return search_listing_ts(exchange, symbol, mid_time, end_time, timeframe)
+    else:
+        return search_listing_ts(exchange, symbol, start_time, mid_time, timeframe)
+    
 def _fetch_candles_ccxt(
     start_ts : int,
     end_ts : int,
     exchange,
     normalized_symbols : List[str],
     candle_size : str,
-    num_candles_limit : int = 100,
-    logger = None,
-    cache_dir : Union[str, None] = None,
-    list_ts_field : Union[str, None] = None
-) -> Dict[str, Union[pd.DataFrame, None]]:
-  ticker = normalized_symbols[0]
-  pd_candles = _fetch_candles(
-              symbol = ticker,
-              exchange = exchange,
-              start_ts = start_ts,
-              end_ts = end_ts,
-              candle_size = candle_size,
-          )
-  return {
-      ticker : pd_candles
-  }
-
-def _fetch_candles(
-    symbol : str,
-    exchange : CcxtExchange,
-    start_ts : int,
-    end_ts : int,
-    candle_size : str = '1d',
     num_candles_limit : int = 100
-):
-    def _fetch_ohlcv(exchange, symbol, timeframe, since, limit, params) -> Union[List, NoReturn]:
-        one_timeframe = f"1{timeframe[-1]}"
-        candles = exchange.fetch_ohlcv(symbol=symbol, timeframe=one_timeframe, since=since, limit=limit, params=params)
-        if candles and len(candles)>0:
-            candles.sort(key=lambda x : x[0], reverse=False)
-
-        return candles
-
-    all_candles = []
-    params = {}
-    this_cutoff = start_ts
-    while this_cutoff<=end_ts:
-        candles = _fetch_ohlcv(exchange=exchange, symbol=symbol, timeframe=candle_size, since=int(this_cutoff * 1000), limit=num_candles_limit, params=params)
-        if candles and len(candles)>0:
-            all_candles = all_candles + [[ int(x[0]), float(x[1]), float(x[2]), float(x[3]), float(x[4]), float(x[5]) ] for x in candles if x[1] and x[2] and x[3] and x[4] and x[5] ]
-
-            record_ts = max([int(record[0]) for record in candles])
-            record_ts_str : str = str(record_ts)
-            if len(record_ts_str)==13:
-                record_ts = int(int(record_ts_str)/1000) # Convert from milli-seconds to seconds
-
-            this_cutoff = record_ts  + 1
-    columns = ['exchange', 'symbol', 'timestamp_ms', 'open', 'high', 'low', 'close', 'volume']
-    pd_all_candles = pd.DataFrame([ [ exchange.name, symbol, x[0], x[1], x[2], x[3], x[4], x[5] ] for x in all_candles], columns=columns)
-    fix_column_types(pd_all_candles)
-    pd_all_candles['pct_chg_on_close'] = pd_all_candles['close'].pct_change()
-    return pd_all_candles
+) -> Dict[str, Union[pd.DataFrame, None]]:
+    logger = logging.getLogger()
+
+    rsp = {}
+
+    exchange.load_markets()
+    
+    num_tickers = len(normalized_symbols)
+    i = 0
+    for ticker in normalized_symbols:
+        @retry(num_attempts=3, pause_between_retries_ms=1000)
+        def _fetch_ohlcv(exchange, symbol, timeframe, since, limit, params) -> Union[List, NoReturn]:
+                one_timeframe = f"1{timeframe[-1]}"
+                candles = exchange.fetch_ohlcv(symbol=symbol, timeframe=one_timeframe, since=since, limit=limit, params=params)
+                if candles and len(candles)>0:
+                    candles.sort(key=lambda x : x[0], reverse=False)
+
+                return candles
+            
+        def _calc_increment(candle_size):
+            increment = 1
+            num_intervals = int(candle_size.replace(candle_size[-1],''))
+            interval_type = candle_size[-1]
+            if interval_type == "m":
+                increment = 60
+            elif interval_type == "h":
+                increment = 60*60
+            elif interval_type == "d":
+                increment = 60*60*24
+            else:
+                raise ValueError(f"Invalid candle_size {candle_size}")
+            return num_intervals * increment
+        
+        logger.info(f"{i}/{num_tickers} Fetching {candle_size} candles for {ticker}.")
+
+        '''
+        It uses a while loop to implement a sliding window to download candles between start_ts and end_ts. 
+        However, start_ts for example can be 1 Jan 2021 for a given ticker. 
+        But if that ticker listing date is 1 Jan 2025, this while loop would waste a lot of time loop between 1 Jan 2021 thru 31 Dec 2024, slowly incrementing this_cutoff += _calc_increment(candle_size).
+        A more efficient way is to find listing date. Start looping from there.
+        '''
+        market = exchange.markets[ticker]
+        this_ticker_start_ts = start_ts
+        if market['created']:
+            this_ticker_start_ts = max(this_ticker_start_ts, int(market['created']/1000))
+
+        all_candles = []
+        params = {}
+        this_cutoff = this_ticker_start_ts
+        while this_cutoff<end_ts:
+            candles = _fetch_ohlcv(exchange=exchange, symbol=ticker, timeframe=candle_size, since=int(this_cutoff * 1000), limit=num_candles_limit, params=params)
+            if candles and len(candles)>0:
+                all_candles = all_candles + [[ int(x[0]), float(x[1]), float(x[2]), float(x[3]), float(x[4]), float(x[5]) ] for x in candles if x[1] and x[2] and x[3] and x[4] and x[5] ]
+
+                record_ts = max([int(record[0]) for record in candles])
+                record_ts_str : str = str(record_ts)
+                if len(record_ts_str)==13:
+                    record_ts = int(int(record_ts_str)/1000) # Convert from milli-seconds to seconds
+                
+                this_cutoff = record_ts  + _calc_increment(candle_size)
+            else:
+                this_cutoff += _calc_increment(candle_size)
+
+        columns = ['exchange', 'symbol', 'timestamp_ms', 'open', 'high', 'low', 'close', 'volume']
+        pd_all_candles = pd.DataFrame([ [ exchange.name, ticker, x[0], x[1], x[2], x[3], x[4], x[5] ] for x in all_candles], columns=columns)
+        fix_column_types(pd_all_candles)
+        pd_all_candles['pct_chg_on_close'] = pd_all_candles['close'].pct_change()
+
+        rsp[ticker] = pd_all_candles
+
+        i+=1
+
+    return rsp
 
 def fetch_deribit_btc_option_expiries(
     market: str = 'BTC'
@@ -404,7 +739,7 @@ def fetch_deribit_btc_option_expiries(
         Dict[str, Dict[str, Union[str, float]]]
     ]
 ]:
-    exchange = deribit()
+    exchange = ccxt.deribit()
     instruments = exchange.public_get_get_instruments({
         'currency': market,
         'kind': 'option',
@@ -456,4 +791,56 @@ def fetch_deribit_btc_option_expiries(
         'index_price' : index_price,
         'by_expiry' : sorted_expiry_data, # type: ignore Otherwise, Error: Type "dict[str, list[tuple[str, float]] | dict[str, Dict[Unknown, Unknown]]]" is not assignable to return type "Dict[str, Dict[str, float] | Dict[str, Dict[str, str | float]]]"
         'by_expiry_and_strike' : expiry_data_breakdown_by_strike
-    }
+    }
+
+def build_pair_candles(
+    pd_candles1 : pd.DataFrame,
+    pd_candles2 : pd.DataFrame,
+    left_columns_postfix : str = "_1",
+    right_columns_postfix : str = "_2"
+) -> pd.DataFrame:
+    min_timestamp_ms1 = int(pd_candles1.iloc[0]['timestamp_ms'])
+    max_timestamp_ms1 = int(pd_candles1.iloc[-1]['timestamp_ms'])
+    min_timestamp_ms2 = int(pd_candles2.iloc[0]['timestamp_ms'])
+    max_timestamp_ms2 = int(pd_candles2.iloc[-1]['timestamp_ms'])
+
+    pd_candles1 = pd_candles1[(pd_candles1.timestamp_ms>=min_timestamp_ms2) & (pd_candles1.timestamp_ms<=max_timestamp_ms2) & (~pd_candles1.timestamp_ms.isna()) ]
+    pd_candles2 = pd_candles2[(pd_candles2.timestamp_ms>=min_timestamp_ms1) & (pd_candles2.timestamp_ms<=max_timestamp_ms1) & (~pd_candles2.timestamp_ms.isna())]
+    assert(pd_candles1.shape[0]==pd_candles2.shape[0])
+
+    pd_candles1['timestamp_ms_gap'] = pd_candles1['timestamp_ms'] - pd_candles1['timestamp_ms'].shift(1)
+    timestamp_ms_gap = pd_candles1.iloc[-1]['timestamp_ms_gap']
+    
+    assert(pd_candles1[~pd_candles1.timestamp_ms_gap.isna()][pd_candles1.timestamp_ms_gap!=timestamp_ms_gap].shape[0]==0)
+    pd_candles1.drop(columns=['timestamp_ms_gap'], inplace=True)
+
+    pd_candles2['timestamp_ms_gap'] = pd_candles2['timestamp_ms'] - pd_candles2['timestamp_ms'].shift(1)
+    timestamp_ms_gap = pd_candles2.iloc[-1]['timestamp_ms_gap']
+    assert(pd_candles2[~pd_candles2.timestamp_ms_gap.isna()][pd_candles2.timestamp_ms_gap!=timestamp_ms_gap].shape[0]==0)
+    pd_candles2.drop(columns=['timestamp_ms_gap'], inplace=True)
+
+    min_timestamp_ms1 = int(pd_candles1.iloc[0]['timestamp_ms'])
+    max_timestamp_ms1 = int(pd_candles1.iloc[-1]['timestamp_ms'])
+    min_timestamp_ms2 = int(pd_candles2.iloc[0]['timestamp_ms'])
+    max_timestamp_ms2 = int(pd_candles2.iloc[-1]['timestamp_ms'])
+    assert(min_timestamp_ms1==min_timestamp_ms2)
+    assert(max_timestamp_ms1==max_timestamp_ms2)
+    assert(pd_candles1.shape[0]==pd_candles2.shape[0])
+
+    if len([ col for col in pd_candles1.columns if col[-2:]==left_columns_postfix ]) == 0:
+        pd_candles1.columns = [str(col) + left_columns_postfix for col in pd_candles1.columns]
+
+    if len([ col for col in pd_candles2.columns if col[-2:]==right_columns_postfix ]) == 0:
+        pd_candles2.columns = [str(col) + right_columns_postfix for col in pd_candles2.columns]
+
+    pd_candles1.reset_index(drop=True, inplace=True)
+    pd_candles2.reset_index(drop=True, inplace=True)
+    pd_candles = pd.concat([pd_candles1, pd_candles2], axis=1)
+    pd_candles['timestamp_ms_gap'] = pd_candles[f'timestamp_ms{left_columns_postfix}'] - pd_candles[f'timestamp_ms{right_columns_postfix}']
+    assert(pd_candles[pd_candles.timestamp_ms_gap!=0].shape[0]==0)
+
+    pd_candles.drop(pd_candles.columns[pd_candles.columns.str.contains('unnamed',case = False)],axis = 1, inplace = True)
+
+    return pd_candles
+
+