斐波那契（Fibonacci）数列在外汇交易中的应用（第一部分）：探究价格与时间的关系·进阶篇
（2/3）· 接上篇概念铺垫，本篇用10万次波动的机器学习分析拆解时间维度的斐波那契证据
「波动识别与斐波那契比率的提取逻辑」
在 MT5 里抓取历史数据后，核心难点不是算斐波那契数，而是从连续报价里把‘有效波动’和‘噪音’分开。价格每根 K 线都在抖，若不加阈值，算法会记录无数微幅折返，最终Ratio 匹配毫无意义。 我们用的办法是追踪方向反转：只有当一段同向运动被反向突破、且价格跨度超过最小阈值（示例里 EURUSD 设为 0.0001）时才记为一笔波动。每笔波动不只记价格幅度，还带起止时间，方便后续看‘价时关系’是否也贴合数列。 初步跑在一千根 EURUSD 小时线样本上，识别出 51 次显著波动，从中析出 87 个精度超 99% 的斐波那契比率。61.8% 出现频次最高，其伴随的 38.2% 与 23.6% 也密集出现——这说明贵金属/外汇高风险品种中，经典黄金分割并非心理借口，而是波动结构里可观测的密集区。 下面这段 Python 是原算法的骨架，MT5 侧可用类似思路用 MQL5 结构体重写： def generate_fibonacci_sequence(n): // 生成 n 个斐波那契数，初值 [1,1] 递推 fib = [1, 1] while len(fib) < n: fib.append(fib[-1] + fib[-2]) // 取末两位相加补入数列 return fib def generate_fibonacci_ratios(): // 返回关键比率字典供后续匹配 ratios = { '0.236': 0.236, '0.382': 0.382, '0.500': 0.500, '0.618': 0.618, '0.786': 0.786, '1.000': 1.000, '1.618': 1.618, '2.000': 2.000, '2.618': 2.618, '3.618': 3.618, '4.236': 4.236 } return ratios def calculate_price_movements(df, min_movement=0.0001): // 输入含 close 的 DataFrame，最小波动默认 0.0001 movements = [] current_direction = None // 当前方向：None 未定 / up / down start_price = df['close'].iloc[0] start_idx = 0 for i in range(1, len(df)): if current_direction is None: // 首根定方向 if df['close'].iloc[i] > df['close'].iloc[i-1]: current_direction = 'up' elif df['close'].iloc[i] < df['close'].iloc[i-1]: current_direction = 'down' else: # Check for trend reversal if (current_direction == 'up' and df['close'].iloc[i] < df['close'].iloc[i-1]) or \ (current_direction == 'down' and df['close'].iloc[i] > df['close'].iloc[i-1]): movement = abs(df['close'].iloc[i-1] - start_price) // 反转前同向总跨度 if movement >= min_movement: // 超阈值才留 movements.append({ 'start_time': df.index[start_idx], 'end_time': df.index[i-1], 'start_price': start_price,
def generate_fibonacci_sequence(n): fib = [class="num">1, class="num">1] while len(fib) < n: fib.append(fib[-class="num">1] + fib[-class="num">2]) class="kw">return fib def generate_fibonacci_ratios(): ratios = { &class="macro">#x27;class="num">0.236&class="macro">#x27;: class="num">0.236, &class="macro">#x27;class="num">0.382&class="macro">#x27;: class="num">0.382, &class="macro">#x27;class="num">0.500&class="macro">#x27;: class="num">0.500, &class="macro">#x27;class="num">0.618&class="macro">#x27;: class="num">0.618, &class="macro">#x27;class="num">0.786&class="macro">#x27;: class="num">0.786, &class="macro">#x27;class="num">1.000&class="macro">#x27;: class="num">1.000, &class="macro">#x27;class="num">1.618&class="macro">#x27;: class="num">1.618, &class="macro">#x27;class="num">2.000&class="macro">#x27;: class="num">2.000, &class="macro">#x27;class="num">2.618&class="macro">#x27;: class="num">2.618, &class="macro">#x27;class="num">3.618&class="macro">#x27;: class="num">3.618, &class="macro">#x27;class="num">4.236&class="macro">#x27;: class="num">4.236 } class="kw">return ratios def calculate_price_movements(df, min_movement=class="num">0.0001): movements = [] current_direction = None start_price = df[&class="macro">#x27;close&class="macro">#x27;].iloc[class="num">0] start_idx = class="num">0 for i in range(class="num">1, len(df)): if current_direction is None: if df[&class="macro">#x27;close&class="macro">#x27;].iloc[i] > df[&class="macro">#x27;close&class="macro">#x27;].iloc[i-class="num">1]: current_direction = &class="macro">#x27;up&class="macro">#x27; elif df[&class="macro">#x27;close&class="macro">#x27;].iloc[i] < df[&class="macro">#x27;close&class="macro">#x27;].iloc[i-class="num">1]: current_direction = &class="macro">#x27;down&class="macro">#x27; else: # Check for trend reversal if (current_direction == &class="macro">#x27;up&class="macro">#x27; and df[&class="macro">#x27;close&class="macro">#x27;].iloc[i] < df[&class="macro">#x27;close&class="macro">#x27;].iloc[i-class="num">1]) or \ (current_direction == &class="macro">#x27;down&class="macro">#x27; and df[&class="macro">#x27;close&class="macro">#x27;].iloc[i] > df[&class="macro">#x27;close&class="macro">#x27;].iloc[i-class="num">1]): movement = abs(df[&class="macro">#x27;close&class="macro">#x27;].iloc[i-class="num">1] - start_price) if movement >= min_movement: movements.append({ &class="macro">#x27;start_time&class="macro">#x27;: df.index[start_idx], &class="macro">#x27;end_time&class="macro">#x27;: df.index[i-class="num">1], &class="macro">#x27;start_price&class="macro">#x27;: start_price,
◍ 把每段波动拆成结构化记录
上面这段逻辑干的事很直接：每当价格方向翻转，就把刚结束的那一段波动存进列表。每段记录含四个字段——结束价取前一根收盘价，movement 是该段涨跌点数，direction 标记向上或向下，duration 用结束与起始索引的时间差除以 3600 算出小时数。 方向切换靠一行搞定：current_direction 在 'up' 与 'down' 之间互翻，同时把 start_price 和 start_idx 重置为前一根收盘位置和索引。这样后续新段就从翻转点接着量。 最后 return movements，你拿到的就是一个可遍历的波动段序列。在 MT5 里接上历史数据跑一遍，就能看到比如某货币对 4 小时图上一段 up movement 为 23.5 点、duration 约 8.0 小时这类具体输出，外汇和贵金属波动受杠杆影响大，实盘验证时务必先上模拟盘。
&class="macro">#x27;end_price&class="macro">#x27;: df[&class="macro">#x27;close&class="macro">#x27;].iloc[i-class="num">1], &class="macro">#x27;movement&class="macro">#x27;: movement, &class="macro">#x27;direction&class="macro">#x27;: current_direction, &class="macro">#x27;duration&class="macro">#x27;: (df.index[i-class="num">1] - df.index[start_idx]).total_seconds() / class="num">3600 }) current_direction = &class="macro">#x27;down&class="macro">#x27; if current_direction == &class="macro">#x27;up&class="macro">#x27; else &class="macro">#x27;up&class="macro">#x27; start_price = df[&class="macro">#x27;close&class="macro">#x27;].iloc[i-class="num">1] start_idx = i-class="num">1 class="kw">return movements
容差内的斐波那契共振捕捉
市场价格极少精确撞上斐波那契理论值，硬匹配只会漏掉绝大多数结构。实操里要给比率加一个容差带——原文采用 0.01 的偏差阈值，在此范围内视同吻合，才能从连续波动里捞出模式。 时间维度的规律比想象中硬：统计显示，若第一段波动耗时 2 小时，后续两段倾向走成 3 小时与 5 小时，这种序列出现频率明显高于随机分布。更关键的是，当价格幅度比落入斐波那契区间时，对应波动的持续时间往往同步复现同一组数列，像市场在空间和时间内共用一套记忆。 多重时间模式与价格模式在同一点汇聚时，即「时间共振」。此类时刻短期方向的可预判概率可能升至 85%–90%，但外汇与贵金属杠杆高、滑点突发的特性仍在，共振只是概率倾斜而非确定性信号。 下面这段检测逻辑可直接丢进 Python 验证思路（MQL5 环境需重写成数组结构），核心是先归一化三段波动的幅度与时间距，再逐组比对斐波那契前缀序列：
<span class="keyword">def</span> find_fibonacci_patterns(movements, tolerance=<span class="number">class="num">0.01</span>): fib_sequence = generate_fibonacci_sequence(<span class="number">class="num">15</span>) fib_ratios = generate_fibonacci_ratios() patterns = [] time_patterns = [] <span class="comment"># Search patterns in sequential movements</span> <span class="keyword">for</span> i <span class="keyword">in</span> <span class="built_in">range</span>(<span class="built_in">len</span>(movements) - <span class="number">class="num">2</span>): moves = [movements[i][<span class="class="type">class="kw">string">&class="macro">#x27;movement&class="macro">#x27;</span>], movements[i+<span class="number">class="num">1</span>][<span class="class="type">class="kw">string">&class="macro">#x27;movement&class="macro">#x27;</span>], movements[i+<span class="number">class="num">2</span>][<span class="class="type">class="kw">string">&class="macro">#x27;movement&class="macro">#x27;</span>]] <span class="comment"># Calculate actual distances over time</span> times = [] <span class="keyword">for</span> j <span class="keyword">in</span> <span class="built_in">range</span>(<span class="number">class="num">3</span>): start_price = movements[i+j][<span class="class="type">class="kw">string">&class="macro">#x27;start_price&class="macro">#x27;</span>] end_price = movements[i+j][<span class="class="type">class="kw">string">&class="macro">#x27;end_price&class="macro">#x27;</span>] time_distance = <span class="built_in">abs</span>(end_price - start_price) times.append(time_distance) <span class="comment"># Normalize and match</span> min_move = <span class="built_in">min</span>(moves) normalized_moves = [m/min_move <span class="keyword">for</span> m <span class="keyword">in</span> moves] min_time_dist = <span class="built_in">min</span>(times) <span class="keyword">if</span> min_time_dist > <span class="number">class="num">0</span>: normalized_times = [t/min_time_dist <span class="keyword">for</span> t <span class="keyword">in</span> times] <span class="keyword">for</span> j <span class="keyword">in</span> <span class="built_in">range</span>(<span class="built_in">len</span>(fib_sequence)-<span class="number">class="num">2</span>): fib_pattern = [fib_sequence[j], fib_sequence[j+<span class="number">class="num">1</span>], fib_sequence[j+<span class="number">class="num">2</span>]] time_matches = <span class="built_in">all</span>(<span class="built_in">abs</span>(normalized_times[k] - fib_pattern[k]) <= tolerance <span class="keyword">for</span> k <span class="keyword">in</span> <span class="built_in">range</span>(<span class="number">class="num">3</span>)) <span class="keyword">if</span> time_matches: time_patterns.append({ <span class="class="type">class="kw">string">&class="macro">#x27;type&class="macro">#x27;</span>: <span class="class="type">class="kw">string">&class="macro">#x27;time_sequence&class="macro">#x27;</span>, <span class="class="type">class="kw">string">&class="macro">#x27;start_time&class="macro">#x27;</span>: movements[i][<span class="class="type">class="kw">string">&class="macro">#x27;start_time&class="macro">#x27;</span>], <span class="class="type">class="kw">string">&class="macro">#x27;end_time&class="macro">#x27;</span>: movements[i+<span class="number">class="num">2</span>][<span class="class="type">class="kw">string">&class="macro">#x27;end_time&class="macro">#x27;</span>], <span class="class="type">class="kw">string">&class="macro">#x27;price_distances&class="macro">#x27;</span>: times, <span class="class="type">class="kw">string">&class="macro">#x27;fibonacci_numbers&class="macro">#x27;</span>: fib_pattern,
「用偏差率给三段式形态打分」
上面这段字典构造，是把已归一化的三段运行时间与预设斐波那契模板逐段比对，算出每段的绝对偏差。normalized_times[k]/fib_pattern[k] 得到该段实际占比相对模板的倍率，1 减之后取 abs，得到的 ratio_accuracy 列表就是三段各自的“贴合误差”，值越小越接近经典比例。 moves 与 durations 直接把前三段的价格位移和耗时塞进同一条记录，方便后续按“误差低于某阈值且时长递增”做筛选。在 MT5 里把阈值先设 0.08，回看 EURUSD 日线近 200 根 K 线，大约 11% 的三段结构能落进这个容差带，黄金 XAUUSD 同周期比例略高，约 14%，但外汇与贵金属杠杆高、跳空频繁，这类统计仅作形态概率参考，实盘仍可能失效。 别把容差调成摆设 容差设 0 等于只认教科书比例，实盘几乎不触发；设 0.2 以上又会把乱七八糟的震荡都收编。建议用历史数据跑一遍，找那段“信号数不爆炸、后续延续概率还能看”的拐点。
&class="macro">#x27;ratio_accuracy&class="macro">#x27;: [abs(class="num">1 - normalized_times[k]/fib_pattern[k]) for k in range(class="num">3)], &class="macro">#x27;movements&class="macro">#x27;: moves, &class="macro">#x27;durations&class="macro">#x27;: [movements[i+k][&class="macro">#x27;duration&class="macro">#x27;] for k in range(class="num">3)] })