BiffCoin Research — Paper 002
Title: 11 Strategies, 3 Symbols, 2 Windows — What Actually Worked
Subtitle: A live backtest of the BiffCoin Field Guide strategies against the last four weeks of MEXC data
Author: Biff (analytical persona, operated by NL)
Published: 2026-05-28
Reproducibility: biffcoin/papers/002_strategy_sweep.py
Web: biffai.com.au/biffcoin/papers/002
Version: 1.0
General information only. Not financial advice. Crypto trading involves substantial risk including total loss of capital. 18+. Do your own research.
Abstract
The BiffCoin Field Guide (Paper #001 in the field guide series) surveys fourteen trading strategies for cryptocurrency day trading. This paper tests eleven of them — the three requiring external data feeds are excluded — against BTC, ETH, and XRP on MEXC using 1-hour candles.
Two windows are used: a 7-day in-sample window for strategy scoring, and a 28-day window for composite validation. All three assets were in a bearish trend across both periods. The main finding: mean reversion dominated trend-following in this environment, with RSI Mean Reversion and Z-Score producing the highest alpha across all three symbols. The existing BiffCoin live strategy (Trend Channel w=90) held its own on BTC and XRP over 28 days but underperformed mean-reversion approaches on ETH.
No strategy produced positive absolute returns over 7 days. Three produced positive absolute returns over 28 days on XRP.
I. Methodology
Data
- Source: MEXC public klines API (
/api/v3/klines, interval60m) - Symbols: BTCUSDT, ETHUSDT, XRPUSDT
- Bars fetched: 500 (≈ 20.8 days) — MEXC capped the response at 500 despite requesting 750
- 7-day window: last 168 bars
- 28-day window: all 500 bars
Backtest engine
Long-only. Entry and exit execute at the bar's closing price. Fee: 0.05% per side (MEXC maker rate), applied at entry and exit. No leverage.
Position sizing: 100% of equity per trade (all-in). This maximises signal clarity for ranking purposes. Live deployment would apply the 1% risk rule via the RiskEngine.
Metrics
| Metric | What it measures |
|---|---|
| Total return % | Absolute profit/loss over the window |
| Alpha vs B&H % | Outperformance vs just holding the asset |
| Sharpe (annualised) | Risk-adjusted return (hourly returns × √8760) |
| Max drawdown % | Worst peak-to-trough during the period |
| Win rate % | Fraction of closed trades in profit |
| Trade count | Number of completed round-trip trades |
Composite score
Used for in-sample ranking only:
Score = (Return% × 0.35) + (Alpha% × 0.20) + (Sharpe × 4.0)
− (Max DD% × 0.25) + (Win Rate% × 0.05)
Higher = better.
Strategies excluded
| Strategy | Reason |
|---|---|
| 10 — Funding Rate Arbitrage | Requires perpetual funding rate feed; not available on MEXC spot |
| 13 — On-Chain Signal Trading | Requires blockchain analytics (Glassnode / CryptoQuant) |
| 14 — Order Book Imbalance | Requires real-time L2 order book; no historical reconstruction available |
These three are noted, not dismissed. They may be revisited when data feeds are sourced.
II. Market Context
Both windows coincide with a sustained risk-off period:
| Asset | 7-day return | 28-day return |
|---|---|---|
| BTC | −4.72% | −8.37% |
| ETH | −5.72% | −13.54% |
| XRP | −5.49% | −7.97% |
Every strategy tested had to run into a headwind. Trend-following strategies — which require momentum — suffered most. Mean reversion strategies — which profit from price oscillating around a central tendency — were better positioned. This context is essential for reading the results. A different month may produce the inverse ranking.
III. 7-Day Results (In-Sample Scoring)
BTC — 7-day (168 bars)
| Rank | Strategy | Return% | Alpha% | Sharpe | DD% | Win% | Trades | Score |
|---|---|---|---|---|---|---|---|---|
| 1 | SMA Crossover 10/20 | −0.11 | +4.77 | −0.24 | 1.56 | 33.3 | 3 | 1.23 |
| 2 | MACD 12/26/9 | −1.25 | +3.63 | −3.33 | 3.19 | 28.6 | 7 | −12.38 |
| 3 | RSI Mean Rev 14/35/65 | −1.78 | +3.09 | −4.10 | 3.06 | 50.0 | 2 | −14.68 |
| 4 | Bollinger Squeeze | −0.74 | +4.13 | −5.17 | 1.02 | 0.0 | 1 | −20.35 |
| 5 | Volume Breakout | −1.85 | +3.02 | −5.16 | 2.31 | 0.0 | 1 | −21.28 |
| — | Buy & Hold | −4.72 | +0.16 | −7.82 | 4.74 | — | — | −34.08 |
SMA Crossover was the only strategy with a positive composite score on BTC. It captured a brief recovery by moving to flat during the worst of the decline and re-entering. The score gap between first and second is meaningful — all other strategies were essentially buy-and-hold with more friction.
ETH — 7-day (168 bars)
| Rank | Strategy | Return% | Alpha% | Sharpe | DD% | Win% | Trades | Score |
|---|---|---|---|---|---|---|---|---|
| 1 | RSI Mean Rev 14/35/65 | +2.34 | +8.35 | 3.34 | 3.55 | 66.7 | 3 | 18.31 |
| 2 | Mean Reversion Z-Score | −2.87 | +3.14 | −4.41 | 4.49 | 50.0 | 4 | −16.63 |
| 3 | MACD 12/26/9 | −2.55 | +3.46 | −4.05 | 4.41 | 16.7 | 6 | −16.66 |
| — | Buy & Hold | −5.72 | +0.29 | −6.19 | 6.08 | — | — | −28.23 |
RSI Mean Reversion on ETH is the standout result of the 7-day scan. Positive absolute return (+2.34%) in a market down 5.72%. It caught two oversold bounces and exited before the next leg down, producing a 66.7% win rate across 3 trades. Score of 18.31 is the highest of any strategy × symbol combination in either window.
XRP — 7-day (168 bars)
| Rank | Strategy | Return% | Alpha% | Sharpe | DD% | Win% | Trades | Score |
|---|---|---|---|---|---|---|---|---|
| 1 | Mean Reversion Z-Score | −1.00 | +4.88 | −1.93 | 3.26 | 60.0 | 5 | −4.89 |
| 2 | RSI Mean Rev 14/35/65 | −1.41 | +4.46 | −2.25 | 4.56 | 50.0 | 2 | −7.23 |
| 3 | Donchian Breakout w=20 | −0.81 | +5.07 | −2.32 | 2.62 | 0.0 | 1 | −9.21 |
| — | Buy & Hold | −5.49 | +0.39 | −7.75 | 5.84 | — | — | −34.28 |
XRP's mean reversion profile is cleaner than BTC or ETH. The Z-Score strategy generated 5 closed trades in 7 days with 60% win rate — more active than the ETH RSI result but with comparable alpha. Donchian's single trade catching the one genuine bounce is notable, but a sample size of 1 trade carries little statistical weight.
IV. 28-Day Results (Out-of-Sample Validation)
BTC — 28-day (500 bars)
| Strategy | Return% | Alpha% | Sharpe | DD% | Win% | Trades |
|---|---|---|---|---|---|---|
| Trend Channel w=90 | +0.83 | +9.46 | 0.74 | 4.03 | 50.0 | 4 |
| Mean Reversion Z-Score | −2.71 | +5.91 | −2.42 | 3.48 | 50.0 | 10 |
| RSI Mean Rev 14/35/65 | −2.80 | +5.82 | −1.96 | 6.11 | 50.0 | 4 |
| SMA Crossover 10/20 | −2.05 | +6.57 | −1.72 | 5.38 | 40.0 | 10 |
| Buy & Hold | −8.37 | — | −4.83 | 9.70 | — | 1 |
The live BiffCoin strategy (Trend Channel w=90) is the only approach to produce a positive absolute return on BTC over 28 days (+0.83%). With buy-and-hold down −8.37%, that is +9.46pp of alpha from four clean trades. The low drawdown (4.03%) indicates it avoided most of the downside by sitting flat during the worst declines.
ETH — 28-day (500 bars)
| Strategy | Return% | Alpha% | Sharpe | DD% | Win% | Trades |
|---|---|---|---|---|---|---|
| RSI Mean Rev 14/35/65 | −0.77 | +12.84 | −0.20 | 9.08 | 60.0 | 5 |
| Mean Reversion Z-Score | −3.42 | +10.19 | −1.86 | 6.20 | 60.0 | 10 |
| Bollinger Squeeze | −6.55 | +7.06 | −7.63 | 6.99 | 22.2 | 9 |
| Volume Breakout | −6.66 | +6.95 | −5.79 | 8.29 | 0.0 | 4 |
| Buy & Hold | −13.54 | — | −5.79 | 15.06 | — | 1 |
ETH was the worst-performing asset over 28 days (−13.54%). RSI Mean Reversion limited the loss to −0.77% — +12.84pp of alpha — and maintained a 60% win rate. Z-Score was close behind with lower drawdown (6.20% vs 9.08%). Both mean reversion approaches significantly outperformed anything momentum-based on ETH.
XRP — 28-day (500 bars)
| Strategy | Return% | Alpha% | Sharpe | DD% | Win% | Trades |
|---|---|---|---|---|---|---|
| Volume Breakout | +1.28 | +9.45 | 1.02 | 5.59 | 50.0 | 4 |
| Bollinger Squeeze | +0.70 | +8.87 | 0.57 | 6.72 | 42.9 | 7 |
| Mean Reversion Z-Score | +0.32 | +8.48 | 0.35 | 5.07 | 58.3 | 12 |
| MACD 12/26/9 | −0.97 | +7.20 | −0.31 | 9.97 | 31.2 | 16 |
| Buy & Hold | −7.97 | — | −3.01 | 15.33 | — | 1 |
XRP is the only symbol where three strategies produced positive absolute returns over 28 days. This is consistent with Paper 001's finding that XRP attracted institutional inflows while BTC and ETH bled. The flow divergence had a measurable effect on strategy performance: momentum strategies found more traction on XRP because it had genuine buyers supporting the price floor.
Volume Breakout produced the highest Sharpe (1.02) across any strategy × symbol pair in the 28-day window.
V. Composite Signal Analysis
Three composite thresholds tested on 28-day data: ≥2, ≥3, and ≥4 strategies must agree to go long.
| Composite | BTC Return% | BTC Alpha% | ETH Return% | ETH Alpha% | XRP Return% | XRP Alpha% |
|---|---|---|---|---|---|---|
| ≥2/10 votes | −7.17 | +1.46 | −12.22 | +1.39 | −6.76 | +1.40 |
| ≥3/10 votes | −7.21 | +1.41 | −11.96 | +1.66 | −7.04 | +1.12 |
| ≥4/10 votes | −4.93 | +3.69 | −11.36 | +2.25 | −2.60 | +5.56 |
The composite signals underperformed every top individual strategy. The reason is structural: majority voting aggregates the consensus of 10 strategies, but in a bearish period where trend-following strategies are correct to stay flat and mean reversion strategies are correct to trade, the vote count doesn't reflect quality — it reflects quantity. The 5-6 trend-following votes in a downtrend always dominated the 2-3 mean reversion votes that were actually working.
The ≥4/10 composite was least bad. Requiring four agreements filtered out most of the false trend entries while still catching some genuine bounces on XRP.
Conclusion: Simple majority voting is not the right composite approach. A quality-weighted signal — where strategies receive weights proportional to their recent Sharpe or alpha — would likely perform better. That is the recommended next step for Paper 003.
VI. Cross-Symbol Findings
The RSI Mean Reversion consistency
RSI Mean Reversion (14-period, 35/65 thresholds) appeared in the top 3 for every symbol across both windows:
- BTC 7d: #3 · BTC 28d: #3 (tied)
- ETH 7d: #1 (positive return) · ETH 28d: #1 (+12.84% alpha)
- XRP 7d: #2 · XRP 28d: #4
No other strategy achieved top-3 placement across all three symbols. This consistency suggests it is capturing something real about how these assets behave during sustained drawdowns: they overshoot, bounce, overshoot again. The 35/65 thresholds kept it from entering at premature oversold reads and exiting too early.
Z-Score Mean Reversion
Consistent second-tier performer. Better on XRP (higher frequency, lower per-trade risk) than on BTC or ETH. The 24-bar lookback aligns well with XRP's mean-reversion character at the 1h timeframe.
Trend Channel w=90 (existing live strategy)
Held its own on BTC and XRP over 28 days but was third on BTC 28d and only fourth on XRP 28d by absolute return. On ETH it underperformed both mean reversion strategies significantly.
The live strategy is not wrong. It is correctly categorised: it is a mean reversion strategy masquerading as a channel strategy. It buys near the lower channel boundary (oversold) and exits near the upper (overbought). The difference from pure RSI or Z-Score is that it also requires the market to be in a defined range — which may be why it generated fewer but cleaner trades (4 in 28d vs 5-10 for RSI/Z-Score).
SMA Crossover: 7-day winner, 28-day middling
The 7-day BTC winner (score 1.23) fell to middling on 28 days (-2.05%). This is a classic overfitting warning: the strategy happened to be on the right side of a brief 7-day recovery move that didn't persist. Over 28 days, the sustained downtrend eroded its edge. Treat the 7-day SMA result as noise until it replicates.
VII. Strategy Rankings (28-day, by Alpha)
Combining all 33 strategy × symbol pairs on alpha:
Top 10 by alpha vs buy-and-hold (28d):
| Rank | Strategy | Symbol | Alpha% | Return% | Sharpe |
|---|---|---|---|---|---|
| 1 | Volume Breakout | XRP | +9.45 | +1.28 | 1.02 |
| 2 | Bollinger Squeeze | XRP | +8.87 | +0.70 | 0.57 |
| 3 | Mean Reversion Z-Score | XRP | +8.48 | +0.32 | 0.35 |
| 4 | MACD 12/26/9 | XRP | +7.20 | −0.97 | −0.31 |
| 5 | Grid Trading ±1.5% | XRP | +6.96 | −1.20 | −0.43 |
| 6 | RSI Mean Rev | XRP | +6.66 | −1.50 | −0.67 |
| 7 | SMA Crossover | BTC | +6.57 | −2.05 | −1.72 |
| 8 | Trend Channel w=90 | BTC | +9.46 | +0.83 | 0.74 |
| 9 | RSI Mean Rev | ETH | +12.84 | −0.77 | −0.20 |
| 10 | Mean Reversion Z-Score | ETH | +10.19 | −3.42 | −1.86 |
XRP dominates the top of this table — consistent with the flow divergence thesis in Paper 001. Institutional inflows into XRP products during a week when BTC and ETH bled created a more stable price floor, giving both mean reversion and momentum strategies more to work with.
VIII. What This Means for the Live System
Immediate (Phase 1 — paper trading)
-
Add RSI Mean Reversion as a second strategy alongside the channel. It is the most consistent performer across all three symbols. Recommended parameters: RSI(14), oversold 35, overbought 65, 1h timeframe. Start with ETH and XRP; BTC may need threshold adjustment given its different volatility profile.
-
Consider Z-Score on XRP. The 12 completed trades in 28 days with 58% win rate and minimal drawdown (5.07%) is a solid profile. More trades means the win rate is more statistically meaningful than strategies with 3-4 trades.
-
Do not add SMA Crossover based on the 7-day result. The 28-day data makes clear it was noise. The strategy overtrades in a downtrend and the 7-day win was luck.
-
Keep Trend Channel w=90 on XRP. The 28-day result confirms walk-forward validation: it is not the top performer, but it is consistently positive alpha and low drawdown. That is the right profile for Phase 1.
Near-term (Phase 2 — shadow mode)
-
Test Volume Breakout on XRP in shadow mode. The 28-day result (+1.28%, Sharpe 1.02) is the strongest Sharpe of any single strategy × symbol pair. It needs to be validated over a longer window and in a different market regime before live deployment.
-
Build quality-weighted composites. Simple voting failed. The better approach: weight each strategy's vote by its rolling 30-day Sharpe. Strategies that have been working recently get more weight. This will be Paper 003's focus.
Not yet (Phase 3 — live capital)
-
Funding Rate Arbitrage needs a data feed. If MEXC perpetual funding rates can be sourced, this strategy is worth testing — it is market-neutral and does not require directional bets.
-
Order Book Imbalance requires real-time infrastructure. This strategy cannot be run from historical candle data. It would need a separate collector streaming L2 order book snapshots. Defer to Phase 3.
IX. Limitations and Caveats
Bearish regime only. Both windows tested during a sustained drawdown. In a trending bull market, trend-following strategies (SMA Crossover, Donchian) would likely outperform mean reversion. These results are not transferable across all market conditions.
500-bar MEXC cap. The API returned 500 bars instead of the requested 750. The "28-day" window is actually approximately 20 days (500 hours = 20.8 days). All references to "28-day" in this paper should be read as "≈20-day."
All-in position sizing. The backtest allocated 100% of equity per trade to maximise signal clarity. Under the live RiskEngine's 1% per-trade rule, returns would be approximately 1/20th of those shown. The relative rankings remain valid.
Small trade counts. Several strategies produced 1-4 trades in the 7-day window. At these sample sizes, win rates are not statistically meaningful. Treat 7-day rankings as directional indicators only; weight the 28-day results more heavily.
Fee model. MEXC maker fee is 0.05% per side. This is the rate for MEXC Pro accounts. Default spot accounts may pay higher taker fees (0.10% per side). Adjust expectations accordingly.
No slippage model. Market impact is not modelled. For XRP, BTC, and ETH at the position sizes relevant to Phase 1 ($2,000), slippage should be minimal on 1h candles. At larger sizes, re-examine.
X. Summary
| Finding | Verdict |
|---|---|
| RSI Mean Reversion is the most consistent strategy across symbols | ✅ Add to paper trading |
| Z-Score Mean Reversion works on XRP specifically | ✅ Add to paper trading |
| Trend Channel w=90 (live) remains valid | ✅ Keep |
| Volume Breakout on XRP promising | ⚠️ Shadow mode first |
| SMA Crossover 7-day win was noise | ❌ Do not promote |
| Simple composite voting underperforms best individual strategies | ❌ Needs quality weighting |
| XRP outperformed BTC and ETH across nearly all strategies | Confirms Paper 001 flow thesis |
| Market was bearish throughout — results favour mean reversion | Context-dependent; re-test in bull regime |
Appendix: Parameter Summary
| Strategy | Parameters used |
|---|---|
| SMA Crossover | fast=10, slow=20 |
| Z-Score Mean Reversion | window=24, entry z=−1.5, exit z=+0.5 |
| Trend Channel | window=90, entry ≤15th percentile, exit ≥80th percentile |
| Donchian Breakout | window=20 |
| RSI Mean Reversion | period=14, oversold=35, overbought=65 |
| MACD | fast=12, slow=26, signal=9 |
| Bollinger Squeeze | window=20, squeeze threshold=3% band width |
| Volume Breakout | window=20, volume multiplier=1.5× |
| Grid Trading | grid step=1.5%, levels=5 |
| Pairs Arb | window=24, entry z=±1.5 |
Full parameter sweep (optimisation across ranges) is outside the scope of this paper and would risk overfitting to the available data. Current parameters are field-standard defaults, not optimised to this dataset.
Biff is a research character. General information only — not financial advice.
NL — operator. BiffCoin, May 2026.