# -*- coding: utf-8 -*- """ AI旺财 EA 马丁格尔网格参数优化模型 本金 60000 USD | XAUUSD | 1手=100oz | 1point=0.01USD """ import random import math PRICE0 = 4320.0 CONTRACT = 100.0 POINT = 0.01 EQUITY = 60000.0 LEV = 2000.0 STEP = 0.01 # ---------------------------------------------------------------- 网格基础 def ladder(init, mult, n, maxlot=99.0): """按券商最小步进 0.01 规整后的实际手数阶梯""" out, v = [], init for _ in range(n): v = min(v, maxlot) q = round(round(v / STEP) * STEP, 2) out.append(q if q >= STEP else STEP) v = v * mult return out def grid_stats(init, mult, n, dist_pts, maxlot=99.0): L = ladder(init, mult, n, maxlot) d = dist_pts * POINT S = sum(L) dd = sum(L[i] * CONTRACT * (i * d) for i in range(n)) return dict(L=L, S=S, dd=dd, d=d, span=(n - 1) * d, margin=S * CONTRACT * PRICE0 / LEV, be=dd / (S * CONTRACT) if S else 0.0) def solve_init(mult, n, dist_pts, target_dd, maxlot=99.0): """反解：满仓浮亏 = target_dd 时的起手手数""" lo, hi, best = 0.001, 50.0, 0.001 for _ in range(80): mid = (lo + hi) / 2 if grid_stats(mid, mult, n, dist_pts, maxlot)['dd'] <= target_dd: best = mid lo = mid else: hi = mid return max(0.01, round(best / STEP) * STEP) # ---------------------------------------------------------------- 解套难度 def unbias_ratio(mult, n): """保本反弹 / 覆盖区间。越小越容易解套，理论下限 0.5（等手数网格）""" s0 = sum(mult ** i for i in range(n)) s1 = sum((mult ** i) * i for i in range(n)) return (s1 / s0) / (n - 1) def show_unbias(): print('=' * 92) print('[1] 加仓倍数对解套难度的影响 (保本反弹 / 网格覆盖区间)') print(' 该比值与起手手数无关，只由 倍数/档数 决定。下限0.5=等手数网格') print('=' * 92) print('%-10s' % '倍数' + ''.join('%10s' % ('N=%d' % n) for n in (6, 8, 10, 12, 15))) print('-' * 92) for mult in (1.0, 1.1, 1.15, 1.2, 1.25, 1.3, 1.4, 1.5, 1.6): line = '%-10s' % ('%.2f' % mult) for n in (6, 8, 10, 12, 15): line += '%9.1f%%' % (unbias_ratio(mult, n) * 100) print(line) # ---------------------------------------------------------------- 蒙特卡洛 def simulate(init, mult, n, dist_pts, maxlot, tp, sl, n_sim=400, hours=6240, sigma=0.28, swap_per_lot_day=-25.0, seed=42): """ 逐小时 GBM 模拟黄金价格，按网格规则交易 tp: 整体止盈金额(USD) sl: 整体止损金额(USD, 0=不启用) 返回 爆仓率/年化收益/最大回撤 等 """ rng = random.Random(seed) d = dist_pts * POINT L = ladder(init, mult, n, maxlot) dt = 1.0 / hours sq = math.sqrt(dt) drift = -0.5 * sigma * sigma * dt vol = sigma * sq results = [] for _ in range(n_sim): eq = EQUITY peak = eq maxdd = 0.0 price = PRICE0 t = 0 # 网格状态 pos = [] # [(lot, entry_price)] nxt = 0 # 下一档索引 opened = 0 rounds = 0 dead = False while t < hours: price *= math.exp(drift + vol * rng.gauss(0, 1)) t += 1 if not pos: # 新一轮：随机方向首单 if rng.random() < 0.02: # 平均每50小时一次首单机会 pos.append((L[0], price, 1 if rng.random() < 0.5 else -1)) nxt = 1 opened = 1 rounds += 1 continue direction = pos[0][2] # 加仓判定 if nxt < n: last = pos[-1][1] if direction * (price - last) <= -d: pos.append((L[nxt], price, direction)) nxt += 1 opened += 1 # 盈亏 pnl = sum(lot * CONTRACT * direction * (price - e) for lot, e, _ in pos) # 隔夜利息（每24小时计一次） pnl += sum(lot for lot, _, _ in pos) * swap_per_lot_day * (t // 24) / 1.0 * 0 # 见下注 equity = eq + pnl if equity > peak: peak = equity dd = peak - equity if dd > maxdd: maxdd = dd # 保证金检查 (0% stop out) margin = sum(lot for lot, _, _ in pos) * CONTRACT * price / LEV if equity - margin <= 0 or equity <= 0: dead = True break if pnl >= tp: eq += pnl pos = [] nxt = 0 continue if sl > 0 and pnl <= -sl: eq += pnl pos = [] nxt = 0 continue # 未平仓按市价结算 if pos: pnl = sum(lot * CONTRACT * direction * (price - e) for lot, e, _ in pos) eq += pnl results.append(dict(dead=dead, ret=(eq - EQUITY) / EQUITY, maxdd=maxdd / EQUITY, rounds=rounds)) alive = [r for r in results if not r['dead']] dead_rate = len([r for r in results if r['dead']]) / len(results) rets = sorted(r['ret'] for r in results) dds = sorted(r['maxdd'] for r in results) return dict( dead_rate=dead_rate, ret_med=rets[len(rets) // 2] if rets else -1, ret_p10=rets[int(len(rets) * 0.1)] if rets else -1, ret_p90=rets[int(len(rets) * 0.9)] if rets else -1, dd_med=dds[len(dds) // 2] if dds else 1, dd_p90=dds[int(len(dds) * 0.9)] if dds else 1, rounds=sum(r['rounds'] for r in results) / len(results), ) if __name__ == '__main__': show_unbias() print() print('=' * 92) print('[2] 候选方案：在「满仓浮亏<=目标回撤」约束下反解起手手数') print('=' * 92) print('%-34s %7s %7s %9s %7s %8s %7s %9s' % ( '方案', '起手', '满仓', '满仓浮亏', '回撤%', '覆盖USD', '覆盖%', '保本反弹')) print('-' * 92) plans = [ ('原版 1.5/N15/2500pt', 1.5, 15, 2500, 0.4, None), ('P1 保守 1.15/N12/4500pt', 1.15, 12, 4500, 99.0, 0.10), ('P2 稳健 1.20/N12/4500pt', 1.20, 12, 4500, 99.0, 0.15), ('P3 均衡 1.25/N12/4500pt', 1.25, 12, 4500, 99.0, 0.20), ('P4 进取 1.30/N12/4500pt', 1.30, 12, 4500, 99.0, 0.28), ('P5 激进 1.40/N12/4500pt', 1.40, 12, 4500, 99.0, 0.40), ('P6 宽网 1.20/N15/4000pt', 1.20, 15, 4000, 99.0, 0.20), ] built = [] for label, mult, n, dpt, maxlot, tgt in plans: if tgt is None: init = 0.01 else: init = solve_init(mult, n, dpt, EQUITY * tgt, maxlot) g = grid_stats(init, mult, n, dpt, maxlot) print('%-34s %7.2f %7.2f %9.0f %6.1f%% %8.0f %6.2f%% %9.1f' % ( label, init, g['S'], g['dd'], g['dd'] / EQUITY * 100, g['span'], g['span'] / PRICE0 * 100, g['be'])) built.append((label, init, mult, n, dpt, maxlot, g)) print() print('=' * 92) print('[3] 各方案阶梯明细') print('=' * 92) for label, init, mult, n, dpt, maxlot, g in built: print(' %-30s 保证金 $%.0f (%.2f%%)' % (label, g['margin'], g['margin'] / EQUITY * 100)) print(' %s' % ' '.join('%.2f' % x for x in g['L'])) print() print('=' * 92) print('[4] 蒙特卡洛 (年化波动28%, 1年逐小时, 400次, 止盈=本金1%=$600)') print(' A: 不设整体止损 B: 设整体止损=本金12%=$7200') print('=' * 92) print('%-34s %8s %9s %9s %8s | %8s %9s %9s' % ( '方案', 'A爆仓率', 'A收益中位', 'A回撤P90', 'A轮次', 'B爆仓率', 'B收益中位', 'B回撤P90')) print('-' * 92) for label, init, mult, n, dpt, maxlot, g in built: a = simulate(init, mult, n, dpt, maxlot, tp=600.0, sl=0.0, n_sim=250) b = simulate(init, mult, n, dpt, maxlot, tp=600.0, sl=7200.0, n_sim=250) print('%-34s %7.1f%% %8.1f%% %8.1f%% %8.0f | %7.1f%% %8.1f%% %8.1f%%' % ( label, a['dead_rate'] * 100, a['ret_med'] * 100, a['dd_p90'] * 100, a['rounds'], b['dead_rate'] * 100, b['ret_med'] * 100, b['dd_p90'] * 100))