전철: 지난 30번의 K선 최고 가격
아래쪽: 지난 30개의 K 라인의 최저 가격
구간 너비: (올라선 - 하선) / (올라선 + 하선)
범위를 넓히면 a보다 작으면, 가격이 상승 경로를 돌파하고, 오픈 시장을 구매하고, 가격이 하부 경로를 평점으로 떨어집니다.
범위를 넓히면 a보다 작으면 가격이 하향 경로를 돌파하고, 오픈 포션을 팔고, 가격이 경로를 돌파합니다.
이 전략에 관심이 있다면 +V:Irene11229로 문의하세요. (내 홈페이지를 클릭하면 더 많은 전략을 지속적으로 업데이트 할 것이며 주요 거래소의 시장 분석 데이터를 얻을 수 있습니다.)
#!/usr/bin/env python3 # -*- coding: utf-8 -*- import json import time import requests from kumex.client import Trade def check_response_data(response_data): if response_data.status_code == 200: try: d = response_data.json() except ValueError: raise Exception(response_data.content) else: if d and d.get('s'): if d.get('s') == 'ok': return d else: raise Exception("{}-{}".format(response_data.status_code, response_data.text)) else: raise Exception("{}-{}".format(response_data.status_code, response_data.text)) def get_kline(s, r, f, t, timeout=5, is_sandbox=False): headers = {} url = 'https://kitchen.kumex.com/kumex-kline/history' if is_sandbox: url = 'https://kitchen-sdb.kumex.com/kumex-kline/history' uri_path = url data_json = '' p = [] if s: p.append("{}={}".format('symbol', s)) if r: p.append("{}={}".format('resolution', r)) if f: p.append("{}={}".format('from', f)) if t: p.append("{}={}".format('to', t)) data_json += '&'.join(p) uri_path += '?' + data_json response_data = requests.request('GET', uri_path, headers=headers, timeout=timeout) return check_response_data(response_data) class Shock(object): def __init__(self): # read configuration from json file with open('config.json', 'r') as file: config = json.load(file) self.api_key = config['api_key'] self.api_secret = config['api_secret'] self.api_passphrase = config['api_passphrase'] self.sandbox = config['is_sandbox'] self.symbol = config['symbol'] self.resolution = int(config['resolution']) self.valve = float(config['valve']) self.leverage = float(config['leverage']) self.size = int(config['size']) self.trade = Trade(self.api_key, self.api_secret, self.api_passphrase, is_sandbox=self.sandbox) if __name__ == "__main__": shock = Shock() while 1: time_to = int(time.time()) time_from = time_to - shock.resolution * 60 * 35 data = get_kline(shock.symbol, shock.resolution, time_from, time_to, is_sandbox=shock.sandbox) print('now time =', time_to) print('symbol closed time =', data['t'][-1]) if time_to != data['t'][-1]: continue now_price = int(data['c'][-1]) print('closed price =', now_price) # high_track high = data['h'][-31:-1] high.sort(reverse=True) high_track = float(high[0]) print('high_track =', high_track) # low_track low = data['l'][-31:-1] low.sort() low_track = float(low[0]) print('low_track =', low_track) # interval_range interval_range = (high_track - low_track) / (high_track + low_track) print('interval_range =', interval_range) order_flag = 0 # current position qty of the symbol position_details = shock.trade.get_position_details(shock.symbol) position_qty = int(position_details['currentQty']) print('current position qty of the symbol =', position_qty) if position_qty > 0: order_flag = 1 elif position_qty < 0: order_flag = -1 if order_flag == 1 and now_price < low_track: order = shock.trade.create_limit_order(shock.symbol, 'sell', position_details['realLeverage'], position_qty, now_price) print('order_flag == 1,order id =', order['orderId']) order_flag = 0 elif order_flag == -1 and now_price > high_track: order = shock.trade.create_limit_order(shock.symbol, 'buy', position_details['realLeverage'], position_qty, now_price) print('order_flag == -1,order id =', order['orderId']) order_flag = 0 if interval_range < shock.valve and order_flag == 0: if now_price > high_track: order = shock.trade.create_limit_order(shock.symbol, 'buy', shock.leverage, shock.size, now_price) print('now price > high track,buy order id =', order['orderId']) order_flag = 1 if now_price < high_track: order = shock.trade.create_limit_order(shock.symbol, 'sell', shock.leverage, shock.size, now_price) print('now price < high track,sell order id =', order['orderId']) order_flag = -1