The Buy-and-Hold Reality: BTC Rose +155% in 2023 and Beat 35 of 36 Strategies

The benchmark sat in the data the whole year and nobody mentioned it until now. In 2023, BTC/USDT went from about $16,600 to about $42,300 — +154.94% total return on an hourly basis, with a maximum drawdown of -21.74% and a Sharpe of 2.42. Then the lab ran 36 strategies against it, and the embarrassing part is not what the strategies did to each other. Strategy Lab #42.

The head-to-head

The chart below shows every one of the 36 strategies after fees and funding. The red line is buy & hold.

After fees and funding, 35 of 36 strategies lost to doing nothing. The single exception: the linear regression channel (post 17), which returned +199.09% net on AAPL daily data with 106 trades — the only strategy in the lab that beat the passive benchmark, and the only one that also beats it at 25bp (+75.88%... which itself still underperforms holding's +154.94%).

So the trend strategies "worked" — in the weirdest possible way

This is the uncomfortable nuance that the per-post commentary danced around. EMA crossover made +89.38% net. That's a great absolute number. It is also 65 points below holding. The trend rules were long for most of the year; they captured the trend and then handed back the re-entry friction and the whipsaw losses that a continuous position never pays.

Only 8 of 36 strategies beat buy & hold on drawdown. The trend winners did not even do that: EMA crossover's worst drawdown was -22.6%, slightly worse than holding's -21.7%. The drawdown winners are the slow reversion and smooth-trend rules — Aroon (-15.8%), %B (-13.3%), RSI(14) (-16.1%), CCI (-18.5%), TRIX (-18.4%), EMA envelope (-17.0%), DEMA (-20.0%), RSI divergence (-15.9%). The honest framing: these strategies traded return for marginally lower drawdown, and almost all of them overpaid.

The three tiers

The middle tier is where almost the whole lab lives. And it is the honest place to be: on a +155% trend year, "positive and below the benchmark" is the normal outcome of any sane long-biased system. What the 10-strategy bottom tier shows is that even sanity is optional.

The uncomfortable conclusion

A real 2023 strategy comparison is not "which indicator is best?" It is "why trade at all?" Every rule in this lab was competing against the free position that a coin flipped to heads on January 1 would have held all year. Only one beat it. The lab's actual deliverable, in my opinion, is not the winning indicators — it's this chart: the cost of trying to be clever is visible as the distance between the tops of the bars and the red line.

Code

import pandas as pd
from backtest_base import backtest_signal, metrics

df = pd.read_pickle("assets/data/binance_BTCUSDT_1h_2023-01-01_2023-12-31.pkl")
rets = df["close"].pct_change().fillna(0.0)
hold = (1 + rets).cumprod().iloc[-1] - 1
print(f"buy & hold 2023: {hold:+.2%}  "
      f"maxDD {(df['close'] / df['close'].cummax() - 1).min():+.2%}")

Reproduce it

cd blog-drafts/scripts
python gen_special_assets.py   # regenerates this chart + all group tables

Data: Binance public API, hourly OHLCV, 8,735 bars. Buy & hold figure also sits in every post.json under buy_hold.

This is a backtest on historical data, not investment advice. Past performance does not predict future results.