Trend Following Trading Strategy Development - Bitcoin

Author

Rui Carvalho Caseiro

Published

March 17, 2026

Show the code

import pandas as pd
pd.options.display.float_format = '{:.2f}'.format
import numpy as np
import plotly.io as pio
pio.renderers.default = "plotly_mimetype+notebook_connected"

import warnings
warnings.filterwarnings("ignore")

import MyCustomLibrary.main as main
import MyCustomLibrary.backtester as bt
import MyCustomLibrary.backtester_tools as tools
import MyCustomLibrary.features as my_ta

1 System Description & Goals

This study will explore the development of a simple trend-following strategy applied to Bitcoin using a moving-average. The core idea is to buy strength when price trades above a simple moving average and exit the trade when it falls below, capturing one of the most robust principles documented in quantitative finance: markets that exhibit momentum tend to continue trending for some time.

System Goals:

- CAGR    > 20%
- MDD     < 40%
- Calmar  > 1

2 Data

Show the code

from MyCustomLibrary.data import BTC_Daily

BTC_Daily

	BTCUSDT
	Open	High	Low	Close	Adj Close
Date
2010-07-17	0.05	0.05	0.05	0.05	NaN
2010-07-18	0.05	0.09	0.05	0.09	NaN
2010-07-19	0.09	0.09	0.08	0.08	NaN
2010-07-20	0.08	0.08	0.07	0.07	NaN
2010-07-21	0.07	0.08	0.07	0.08	NaN
...	...	...	...	...	...
2026-02-14	68856.98	70481.16	68706.62	69767.62	NaN
2026-02-15	69764.95	70939.29	68052.55	68788.19	NaN
2026-02-16	68782.40	70067.23	67301.59	68843.16	NaN
2026-02-17	68843.09	69201.87	66615.28	67494.22	NaN
2026-02-18	67473.45	68332.81	66767.72	66877.57	NaN

5684 rows × 5 columns

Show the code

BTC_Daily.info()

<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 5684 entries, 2010-07-17 to 2026-02-18
Data columns (total 5 columns):
 #   Column                Non-Null Count  Dtype  
---  ------                --------------  -----  
 0   (BTCUSDT, Open)       5684 non-null   float64
 1   (BTCUSDT, High)       5684 non-null   float64
 2   (BTCUSDT, Low)        5684 non-null   float64
 3   (BTCUSDT, Close)      5684 non-null   float64
 4   (BTCUSDT, Adj Close)  0 non-null      float64
dtypes: float64(5)
memory usage: 266.4 KB

All data is complete, Numbers formatting is correct!

2.0.1 Data prep

Show the code

def df_add_strategy_features(quotes_df: pd.DataFrame, lookbacks: list) -> pd.DataFrame:

    result_df = quotes_df.copy()
    close_series = result_df[('BTCUSDT', 'Close')]

    for lookback in lookbacks:
        result_df.loc[:, ('BTCUSDT', f'SMA{lookback}')] = my_ta.get_SMA(close_series, lookback)
        
    return result_df

Show the code

lookbacks = [5, 10, 20, 40, 80, 160, 320, 640]

main.timer.start()
screened_df = df_add_strategy_features(BTC_Daily, lookbacks)
main.timer.stop()

screened_df.tail()


[*] Total runtime: 4.54 ms

	BTCUSDT
	Open	High	Low	Close	Adj Close	SMA5	SMA10	SMA20	SMA40	SMA80	SMA160	SMA320	SMA640
Date
2026-02-14	68856.98	70481.16	68706.62	69767.62	NaN	68126.65	68355.82	75087.02	83441.93	86389.09	97901.12	101227.58	89101.55
2026-02-15	69764.95	70939.29	68052.55	68788.19	NaN	68125.49	68964.43	74113.07	82818.41	86157.17	97632.77	101170.98	89114.05
2026-02-16	68782.40	70067.23	67301.59	68843.16	NaN	68495.73	68793.21	73100.10	82256.79	85886.22	97371.02	101113.69	89126.62
2026-02-17	68843.09	69201.87	66615.28	67494.22	NaN	68750.21	68614.43	72015.58	81668.46	85588.84	97101.29	101061.00	89136.07
2026-02-18	67473.45	68332.81	66767.72	66877.57	NaN	68354.15	68275.72	71131.38	81077.58	85288.31	96830.37	101011.88	89142.28

3 Backtesting

3.1 Signals generation

Entry: Buy when price crosses over MA

Exits: Exit when price crosses down MA

Show the code

def symbol_above_SMA(symbol_df: pd.DataFrame, sma_lookback: int, row: int)-> bool:
    """ True if yesterday close is above yesterday SMA value."""
    
    return bool(symbol_df['Close'].iloc[row] > symbol_df[f'SMA{sma_lookback}'].iloc[row])


def symbol_below_SMA(symbol_df: pd.DataFrame, sma_lookback: int, row: int)-> bool:
    """ True if yesterday close is below yesterday SMA value."""

    return bool(symbol_df['Close'].iloc[row] < symbol_df[f'SMA{sma_lookback}'].iloc[row])

3.2 Strategy Implementation

Show the code

class TrendStrategy(bt.Strategy):
    
    def init(self,
             sma_lookback: int,
             **kwargs
             ):
        
        self.symbol = 'BTCUSDT'
        self.sma_lookback = sma_lookback

    def next(self, i, record):

        try:
            ###------------Buy conditions------------###
            if not self.has_position(self.symbol) and symbol_above_SMA(self.data[self.symbol], self.sma_lookback, row=i-1):
                
                self.open_long(symbol=self.symbol, price=record[(self.symbol,'Open')])
                      
            ###------------Sell conditions------------###
            elif self.has_position(self.symbol) and symbol_below_SMA(self.data[self.symbol], self.sma_lookback, row=i-1):
                
                self.close(symbol=self.symbol, price=record[(self.symbol,'Open')])
            
        except Exception as e:
            import traceback
            print(traceback.format_exc())

Show the code

lt_result.trades

	symbol	open_date	close_date	bars_held	open_price	close_price	position_size	profit_loss	change_pct	trade_commission	cumulative_return	current_open_positions	why_exit
0	BTCUSDT	2010-07-17	2010-08-27	41	0.05	0.06	2017776.20	29237.58	29.27	229.04	129008.54	1	None
1	BTCUSDT	2010-08-28	2010-09-01	4	0.07	0.06	1982764.00	-9913.82	-7.69	247.85	118846.87	1	None
2	BTCUSDT	2010-10-07	2010-11-30	54	0.06	0.23	1890274.57	315845.98	266.02	553.30	434139.55	1	None
3	BTCUSDT	2010-12-03	2010-12-05	2	0.26	0.21	1700807.23	-83339.55	-19.22	784.07	350015.92	1	None
4	BTCUSDT	2010-12-17	2010-12-18	1	0.25	0.24	1398665.03	-13986.65	-4.00	685.35	335343.93	1	None
5	BTCUSDT	2010-12-21	2010-12-22	1	0.27	0.24	1254715.05	-33877.31	-10.11	636.14	300830.48	1	None
6	BTCUSDT	2010-12-23	2011-03-09	75	0.25	0.87	1202119.80	745314.28	248.00	1346.37	1044798.38	1	None
7	BTCUSDT	2011-03-11	2011-03-12	1	0.93	0.88	1118827.99	-59186.00	-5.67	2028.32	983584.06	1	None
8	BTCUSDT	2011-04-13	2011-07-05	78	0.86	13.86	1142559.83	14853277.85	1511.63	16818.48	15820043.43	1	None

3.3 In Sample Optimization Training Data

It is important that the basic buy-and-sell rules are successful across a range of calculation periods and similar markets. If this fails, then you need to rethink your idea. I don’t believe in adding rules to make the strategy more complex in the hope that one rule will turn a losing idea into a profitable one. It is necessary that the basic concept is robust before moving forward. Then we can try to make it better. — Perry Kaufman

Show the code

df_train, df_test = main.train_test_split(screened_df, test_size=0.20)

optimize = bt.Optimization(TrendStrategy, df_train, df_train[('BTCUSDT', 'Close')], cash=100_000, commission=0.0)

values = [
    lookbacks
    ]

main.timer.start()
optimize.run(values)
main.timer.stop()


[*] Total runtime: 13.30 s

Show the code

main.plot_comparative_graph(optimize.equity_curves())

Visually inspection on the “optimized” strategy equity curves do reveal that the 2 latest tested lookbacks (320 & 640) do underperfom heavily vs shorter lookbacks. Lookbacks to 80 days have overperformed the Benchmark in returns terms.

Show the code

strats_performance =  pd.concat([tools.strategy_performance_metrics(optimize.equity_curves(), steps_per_year=365, rf=0), 
                                 optimize.optimization_trades_stats()], 
                                 axis='index').fillna('-')
strats_performance

	Benchmark	5	10	20	40	80	160	320	640
CAGR %	177.49	212.29	210.23	212.03	225.96	222.57	164.46	74.06	53.10
Ann. Vola %	145.21	76.80	73.80	73.70	76.21	80.26	79.04	75.47	74.32
Max Drawdown %	-93.07	-79.09	-62.99	-64.26	-64.91	-70.28	-84.13	-90.43	-97.79
VaR 5%	-6.73	-3.90	-3.88	-3.95	-4.23	-4.55	-4.64	-5.05	-5.05
Time in Market %	99.52	64.28	61.46	58.34	58.25	58.64	60.01	59.81	66.89
Sharpe	1.22	2.76	2.85	2.88	2.96	2.77	2.08	0.98	0.71
Sortino	2.25	3.46	3.28	3.26	3.25	3.04	2.22	0.98	0.70
Calmar	1.91	2.68	3.34	3.30	3.48	3.17	1.95	0.82	0.54
Efficiency	1.78	3.30	3.42	3.63	3.88	3.80	2.74	1.24	0.79
Nr trades	-	539.00	335.00	209.00	125.00	73.00	40.00	36.00	18.00
Win Rate %	-	38.78	33.23	31.58	32.00	32.88	27.50	25.00	16.67
Profit Factor	-	1.21	1.29	1.37	1.89	1.88	3.47	3.36	3.41
Avg. Win %	-	13.13	22.73	34.94	58.10	100.60	245.71	185.43	1075.39
Avg. Loss %	-	-3.45	-3.79	-3.83	-4.13	-4.20	-5.09	-8.94	-12.95
Reward/Risk	-	3.81	6.00	9.12	14.06	23.95	48.24	20.75	83.02
Expectancy %	-	2.98	5.02	8.41	15.78	30.26	63.88	39.66	168.44
Avg. Bars Held	-	4.00	7.00	12.00	20.00	35.00	68.00	74.00	168.00

The results tell a clear story: trend following works well on Bitcoin, but only when trades are allowed to run long enough to capture major trends. Short holding periods generate many trades but relatively weak performance due to noise and volatility.

The strongest performance appears around 20–80 bars, where risk-adjusted metrics peak (Sharpe near 3, with strong Sortino and Calmar ratios).

The system operates with a low win rate (~30%), but remains profitable due to the large reward-to-risk asymmetry. Losses average around -5%, while winners range from roughly 20% to over 200%, reflecting the classic trend-following principle: cut losses short and let profits run.

As lokback increases, holding periods increase and trade quality improves as expectancy rises sharply.

Show the code

tools.parameter_sensitivity_bar_plot(strats_performance, 'CAGR %')

All lookbacks are profitable, pointing out that these (Entry & Exits) signals are something good to start with. The goal is not to trade/invest on the highest CAGR strategy, but to choose a strategy parameter that is robust to stand the test of time. As stated before, long Lookbacks (160, 320 and 640) have shown to underperform shorter lookbacks.

Show the code

tools.parameter_sensitivity_bar_plot(strats_performance, 'Ann. Vola %')

The trend following filter have managed to cut Bitcoin volatility in half in all of the tested parameters.

Show the code

tools.parameter_sensitivity_bar_plot(strats_performance, 'Max Drawdown %')

In terms of Maximum Drawdowns the trend following filter have delivered the best results on the medium-term 20-80 bars lookback horizons. After 80 bars lookback, the longer the lookback, the longer is the Maximum Drawdown.

Show the code

tools.parameter_sensitivity_bar_plot(strats_performance, 'VaR 5%')

All tested parameters managed to deliever a lower historical VAR5% than Bitcoin Buy & Hold, highlighting Trend Following usefullness for more sophisticated investors. VAR has increased with the increase of lookback period, the longer the lokback is, the bigger the VAR is.

Overall, the results suggest Bitcoin is a highly trend-dominated market, where the strategy’s edge comes from cutting losses quickly and allowing large trends to compound, with the best balance occurring at medium-term holding horizons.

3.4 Out of Sample Validation

Since we discussed earlier that the most reliable lookback horizon was between 20-80 bars, let’s go with the middle figure and test out of sample, including 0.2% cost per trade (Binance Cost Structure).

Show the code

main.timer.start()
os_backtest = bt.Backtest(TrendStrategy, df_test, df_test[('BTCUSDT', 'Close')], cash=100_000, commission=0.002)
os_result = os_backtest.run(sma_lookback=40)
main.timer.stop()


[*] Total runtime: 362.47 ms

Show the code

os_result.stats

	Backtest
Start Period	2023-01-03
End Period	2026-02-18
Duration (Months)	37

CAGR %	56.80
Ann. Vola %	35.02
Max Drawdown %	-24.01
VaR 5%	-2.41
Time in Market %	60.30
Sharpe	1.62
Sortino	2.09
Calmar	2.37
Efficiency	0.94

Nr trades	36.00
Win Rate %	36.11
Profit Factor	3.13
Avg. Win %	17.61
Avg. Loss %	-2.40
Reward/Risk	7.35
Expectancy %	4.83
Avg. Bars Held	18.00

The Out of Sample period spans 37 months from january 2023 until february 2026. In this period, the Trend Following strategy using 40 days lookback managed to deliver a 56% annualized returns figure with roughly 35% annualized volatility leading to a whoping 1.6 raw Sharpe Ratio, 2.1 raw Sortino Ratio and 2.4 Calmar Ratio. Its only been invested ~60% of the time, with a total of 36 trades averaging ~1 trade per month. Consistent with previous In Sample results, its win rate is around 30% with Huge wins and short losses, boosting the Reward/Risk Ratio to ~7.

Show the code

main.plot_strategy_performance(os_result.returns, os_result.benchmark)

Trend Following Strategy has been trailing Buy & Hold for most of the time until just recently due to the recent fall Bitcoin has taken since 10/10 Trump Tariffs first announcement. Highlighting once again the superiorness of Wealth preservation vs Buy & Hold.

3.4.0.1 Returns Analysis

Show the code

tools.strategy_performance_metrics(pd.concat([os_result.returns, os_result.benchmark], axis=1), steps_per_year=365, rf=0)

	Strategy Backtest	Benchmark
CAGR %	56.80	55.86
Ann. Vola %	35.02	47.97
Max Drawdown %	-24.01	-49.74
VaR 5%	-2.41	-3.68
Time in Market %	60.30	100.00
Sharpe	1.62	1.16
Sortino	2.09	1.78
Calmar	2.37	1.12
Efficiency	0.94	0.56

In the out of sample period, Strategy has outperformed Bitcoin Buy & Hold in all fronts like returns, volatility, Maximum Drawdown and Risk Adjusted Returns.

Show the code

tools.metrics_time_window_analysis(pd.concat([os_result.returns, os_result.benchmark], axis=1), lookbacks=[3, 6, 9, 12, 18, 24])

	Strategy Backtest	Benchmark
3-month	48.6%	62.9%
6-month	75.0%	78.1%
9-month	96.6%	93.1%
12-month	92.3%	84.6%
18-month	100.0%	100.0%
24-month	100.0%	100.0%

Up until 9 months, BH proved to be more reliable in terms of wealth preservation than the trend following strategy. On a 12 month rolling scenario, trading strategy delivered a profit 100% of the time vs Buy & Hold ~85%.

Show the code

main.monthly_pos_neg_returns(os_result.returns)

3.4.0.2 Trades Analysis

Show the code

os_result.trades

	symbol	open_date	close_date	bars_held	open_price	close_price	position_size	profit_loss	change_pct	trade_commission	cumulative_return	current_open_positions	why_exit
0	BTCUSDT	2023-01-07	2023-03-01	53	16952.12	23150.93	5.89	36493.61	36.57	472.19	136021.42	1	None
1	BTCUSDT	2023-03-02	2023-03-04	2	23647.02	22362.92	5.74	-7371.58	-5.43	528.26	128121.59	1	None
2	BTCUSDT	2023-03-14	2023-04-22	39	24201.77	27265.89	5.28	16188.80	12.66	543.84	143766.54	1	None
3	BTCUSDT	2023-04-28	2023-05-02	4	29481.01	28087.18	4.87	-6783.60	-4.73	560.35	136422.59	1	None
4	BTCUSDT	2023-05-03	2023-05-08	5	28680.49	28450.46	4.75	-1092.02	-0.80	542.42	134788.16	1	None
5	BTCUSDT	2023-05-29	2023-05-30	1	28075.59	27745.12	4.79	-1583.38	-1.18	534.91	132669.87	1	None
6	BTCUSDT	2023-06-20	2023-07-25	35	26841.66	29178.97	4.93	11529.51	8.71	552.68	143646.70	1	None
7	BTCUSDT	2023-09-19	2023-09-25	6	26760.85	26253.78	5.36	-2716.45	-1.89	568.01	140362.24	1	None
8	BTCUSDT	2023-09-26	2023-09-27	1	26294.76	26209.50	5.33	-454.21	-0.32	559.42	139348.61	1	None
9	BTCUSDT	2023-09-28	2024-01-13	107	26355.81	42799.45	5.28	86767.34	62.39	729.82	225386.14	1	None
10	BTCUSDT	2024-01-30	2024-02-01	2	43300.23	42569.76	5.19	-3794.62	-1.69	892.16	220699.36	1	None
11	BTCUSDT	2024-02-02	2024-02-05	3	43077.64	42577.62	5.11	-2556.63	-1.16	875.92	217266.80	1	None
12	BTCUSDT	2024-02-07	2024-04-03	56	43090.02	65446.67	5.03	112500.83	51.88	1092.33	328675.30	1	None
13	BTCUSDT	2024-04-04	2024-04-13	9	65975.70	67188.38	4.97	6029.22	1.84	1324.14	333380.39	1	None
14	BTCUSDT	2024-05-16	2024-06-14	29	66256.11	66747.57	5.02	2467.94	0.74	1335.80	334512.54	1	None
15	BTCUSDT	2024-07-16	2024-08-03	18	64784.42	61414.81	5.15	-17364.16	-5.20	1300.65	315847.73	1	None
16	BTCUSDT	2024-08-24	2024-08-28	4	64103.87	59507.93	4.92	-22599.59	-7.17	1215.67	292032.47	1	None
17	BTCUSDT	2024-09-14	2024-09-17	3	60569.12	58192.51	4.81	-11435.89	-3.92	1142.93	279453.65	1	None
18	BTCUSDT	2024-09-18	2024-10-10	22	60309.00	60581.93	4.62	1262.15	0.45	1118.11	279597.69	1	None
19	BTCUSDT	2024-10-12	2024-12-23	72	62444.62	95099.39	4.47	145920.91	52.29	1408.00	424110.61	1	None
20	BTCUSDT	2024-12-25	2024-12-27	2	98675.91	95704.98	4.29	-12743.65	-3.01	1667.57	409699.39	1	None
21	BTCUSDT	2025-01-04	2025-01-08	4	98106.99	96924.16	4.17	-4929.69	-1.21	1625.67	403144.03	1	None
22	BTCUSDT	2025-01-16	2025-02-03	18	100505.30	97681.10	4.00	-11305.72	-2.81	1586.75	390251.57	1	None
23	BTCUSDT	2025-02-04	2025-02-05	1	101398.72	97878.01	3.84	-13523.06	-3.47	1530.84	375197.67	1	None
24	BTCUSDT	2025-04-13	2025-04-14	1	85279.47	83694.52	4.39	-6959.25	-1.86	1483.88	366754.54	1	None
25	BTCUSDT	2025-04-15	2025-06-06	52	84539.70	101574.37	4.33	73753.20	20.15	1611.60	438896.15	1	None
26	BTCUSDT	2025-06-07	2025-06-15	8	104390.65	105464.84	4.20	4507.29	1.03	1761.10	441642.35	1	None
27	BTCUSDT	2025-06-17	2025-06-18	1	106794.12	104602.07	4.13	-9047.02	-2.05	1744.95	430850.38	1	None
28	BTCUSDT	2025-06-26	2025-07-02	6	107375.07	105703.10	4.00	-6695.51	-1.56	1706.57	422448.30	1	None
29	BTCUSDT	2025-07-03	2025-08-02	30	108845.02	113320.39	3.87	17335.12	4.11	1721.09	438062.33	1	None
30	BTCUSDT	2025-08-04	2025-08-07	2	114223.92	115030.05	3.83	3085.44	0.71	1754.92	439392.85	1	None
31	BTCUSDT	2025-08-08	2025-08-19	11	117505.50	116241.86	3.73	-4715.75	-1.08	1744.63	432932.47	1	None
32	BTCUSDT	2025-09-11	2025-09-23	12	113961.43	112757.48	3.79	-4564.61	-1.06	1719.14	426648.71	1	None
33	BTCUSDT	2025-09-30	2025-10-11	11	114396.52	113236.43	3.72	-4318.00	-1.01	1694.55	420636.15	1	None
34	BTCUSDT	2025-10-27	2025-10-28	1	114479.85	114129.09	3.67	-1286.25	-0.31	1676.61	417673.28	1	None
35	BTCUSDT	2026-01-03	2026-01-21	18	89945.05	88326.51	4.63	-7500.96	-1.80	1652.36	408519.97	1	None

At a first glance, it appears to be several trades lasting only 1 day, adding up unncessary Transaction Costs due to fakeouts.

Show the code

tools.metrics_trades(os_result.trades, full_metrics=True)

	Trades Analysis
Nr trades	36.00
Win Rate %	36.11
Profit Factor	3.13
Avg. Win %	17.61
Avg. Loss %	-2.40
Reward/Risk	7.35
Expectancy %	4.83
Avg. Bars Held	18.00
Commissions	43979.81
Max Bars Held	107.00
Min Bars Held	1.00
Best Trade %	62.39
Worst Trade %	-7.17
Avg. Open Positions	1.00
Max Open Positions	1.00

Show the code

tools.plot_trades_signals(df_test['BTCUSDT'][['Close', 'SMA40']], os_result.trades)

As stated previously, there are consolidation periods that lead to several in and out trades in a short period of time, adding up transaction costs.

Show the code

clipped_returns = main.cap_series(os_result.trades['change_pct'], percentile=0.005)
ax2 = clipped_returns.plot.hist(bins=50, title="Trades Return Distribution", figsize=(10, 5))

Strategy is working as intended, cutting losses short and letting profits run.

Show the code

ax1 = os_result.trades['bars_held'].plot.hist(bins=30, title="Bars Held Distribution", figsize=(10, 5))
ax1.axvline(os_result.trades['bars_held'].mean(), color='red', linestyle='dashed', linewidth=2)
print("Avg. Bars Held:", int(os_result.trades['bars_held'].mean()))

Avg. Bars Held: 18

Average trade lasted 18 days, there are several very short trades, this may not be ideal.

Show the code

import plotly.express as px

ax2 = px.scatter(os_result.trades, 
                 x='bars_held', 
                 y='change_pct', 
                 width=900,
                 height=500, 
                 trendline="ols", 
                 trendline_color_override="black", 
                 title='Trade duration VS Trade Profit',
                 template='seaborn'
                 )

ax2.show()

Trend is clear, the longer the trade is the more profit it delivers.

3.4.0.3 Statistical Analysis

Show the code

main.correlation_map(pd.concat([os_result.returns, os_result.benchmark], axis=1))

The tested Trend Following Strategy as a 0.7 correlation with Bitcoin Buy & Hold.

3.4.0.4 Is it significantly different from 0?

Show the code

main.compute_t_test(os_result.returns)

P-value: 0.009881528305754408 Reject the null hypothesis (H₀). The strategy average return is significantly different from zero.

The out-of-sample returns exhibit a statistically significant positive mean with a p-value of 0.0099. This suggests that the strategy’s expected return is unlikely to be zero.

3.4.0.5 Does it pass Monte Carlo Permutation?

This will shuffle the given equity curve returns in random order, all output equity curves start and end at the same place, only the path will vary, to obtain Confidence Intervals for Max Draw Down Figures.

Show the code

mcpyt = main.monte_carlo_permutation_yt(os_result.returns, nr_simulations=1000)

Show the code

mcpyt.tail()

	0	1	2	3	4	5	6	7	8	9	...	990	991	992	993	994	995	996	997	998	999
2026-02-14	4.12	4.15	3.66	3.95	4.23	3.45	4.08	4.04	4.11	4.19	...	3.99	4.17	4.06	4.11	3.76	4.16	4.02	3.99	4.24	4.09
2026-02-15	4.18	4.10	3.70	3.97	4.23	3.54	4.06	4.04	4.09	4.19	...	4.06	4.03	4.06	4.07	3.93	4.22	4.12	3.87	4.26	4.01
2026-02-16	4.24	4.05	3.82	4.05	4.23	3.54	4.09	4.09	4.09	4.19	...	4.06	4.12	4.06	4.07	4.00	4.20	4.16	4.07	4.11	4.01
2026-02-17	4.24	4.07	3.92	4.05	4.09	3.73	4.09	4.09	4.09	4.09	...	4.01	4.14	4.09	4.07	4.00	4.23	4.16	4.09	4.09	3.98
2026-02-18	4.09	4.09	4.09	4.09	4.09	4.09	4.09	4.09	4.09	4.09	...	4.09	4.09	4.09	4.09	4.09	4.09	4.09	4.09	4.09	4.09

5 rows × 1000 columns

Show the code

main.mcpyt_analysis(mcpyt, os_result.returns, ci=95)

The observed maximum drawdown lies within the 95% Monte Carlo confidence interval derived from return permutations. The p-value of 0.17 suggests that the realized drawdown is not statistically different from what would be expected under random ordering of returns. While the realized path exhibits somewhat smaller drawdowns than the average randomized path, the result does not provide strong statistical evidence that the strategy benefits from favorable return sequencing.

4 Final Thoughts

The backtests indicate that even a minimal rule-based approach can produce attractive risk-adjusted performance in a strongly trending asset such as Bitcoin, aligning with the original system objectives of high CAGR and acceptable drawdowns.

A future suggestion improvement is to look at reducing trade fakeouts, and analyzing several assets.