Skip to content

8-regime Walk-Forward

Cross-regime walk-forward test of HonestTrend15mDry on BTC/USDT + ETH/USDT (15m timeframe) across 8 fundamentally distinct market regimes, using data pulled from data.binance.vision going back to 2017-08-17 (Binance USDT first day).

Last run: 2026-04-21 (post pyramid adoption) Strategy: HonestTrend15mDry (EMA 94/139 + ADX 18 + FnG gate + pyramid winners) Pairs: BTC/USDT, ETH/USDT Timeframe: 15m Capital: 10,000 USDT simulated, stake_amount=1500, max_open_trades=3 Fees: 0.1% (realistic worst case)

Results — BTC/USDT + ETH/USDT (core pair)

Section titled “Results — BTC/USDT + ETH/USDT (core pair)”
#WindowRegimeTradesTotal %Pass?
W12018-01 → 2018-122018 crash (BTC −72%)53−14.43%No
W22019-01 → 2019-12Accumulation / recovery70+9.57%Yes
W32020-01 → 2020-12COVID crash + recovery60+46.88%Yes
W42021-01 → 2021-12Bull top (BTC $69K)70+34.77%Yes
W52022-01 → 2022-12LUNA + FTX (BTC −64%)65−4.83%No
W62023-01 → 2023-12Recovery / chop78+4.72%Yes
W72024-01 → 2024-12ETF bull + halving63+44.69%Yes
W82025-01 → 2026-04Present88−4.00%No
Total547avg +14.68%5/8

Profitable windows: 5 / 8 (threshold: ≥ 5/8 = statistical edge)

Results — BTC/ETH/BNB/XRP (4-pair extension, pyramid enabled)

Section titled “Results — BTC/ETH/BNB/XRP (4-pair extension, pyramid enabled)”
#WindowTradesTotal %Pass?
W12018 crash111+13.00%Yes
W22019 accumulation154−0.64%No
W32020 COVID110+59.88%Yes
W42021 bull top125+111.96%Yes
W52022 LUNA/FTX128−21.98%No
W62023 recovery157+3.81%Yes
W72024 ETF bull119+63.71%Yes
W82025 present178−1.36%No
Total1082avg +28.5%5/8

Profitable windows: 5 / 8 — pyramid + stake=1500 even flipped the 2018 window to positive (+13%).

New vs old config comparison (core insight of this rerun)

Section titled “New vs old config comparison (core insight of this rerun)”
MetricOld (stake unlimited, no pyramid)New (stake=1500, pyramid enabled)Change
Full-history total profit+954%+195%halved in absolute (per-trade capital halved)
Full-history max DD26.1%16.9%−9.2 ppt
BTC+ETH W1 2018 loss−30.6%−14.4%halved
BTC+ETH+BNB+XRP W1 2018−23.6%+13.0%flipped positive
2-pair profitable windows6/85/8−1 (W8 flipped negative)
4-pair profitable windows4/85/8+1

Fundamental shift: pyramid + fixed stake converts the strategy from “high return high risk” to “medium return low risk”.

Biggest value: max DD dropped from 26% → 17%, meaning under the risk_manager 20% kill-switch threshold the strategy will almost never be retired. The cost is smaller absolute per-trade return — if you want the old aggressiveness back, raise stake_amount to a larger value (e.g. 2500-3000).

Strategy is profitable in 6 different market structures:

  • Strong bull (2020 COVID +97%, 2024 ETF bull +89%, 2021 top +60%)
  • Accumulation (2019 +9.6%)
  • Chop / slow recovery (2023 +4%, 2025 +2.3%)

After expanding to 4 pairs, W2 (2019) and W8 (2025) flipped from positive to negative, dragging the profitable ratio from 6/8 down to 4/8. Reasons:

  • BNB is highly correlated with BTC but has worse signal quality: Binance’s ecosystem token, price is influenced by exchange actions (buyback/burn/new listing announcements), trend signals get repeatedly interrupted by non-market factors
  • XRP is the regulatory coin: during the 2020-2023 SEC lawsuit, price drivers were non-technical, trend-following was guaranteed to get hit
  • Over-diversification: with max_open_trades=3, XRP/BNB steal slots that should have gone to BTC/ETH

No — losses concentrated in “back-to-back crash years”

Section titled “No — losses concentrated in “back-to-back crash years””
  • 2018 (−30.6%): BTC annual −72%, any trend strategy gets whipsawed repeatedly
  • 2022 (−8.3%): LUNA + FTX double blow-up, whole year down without sustained up-legs
  • This environment needs a regime filterHonestTrend1mMTF uses 4h EMA as a gate, does not open positions in clearly downtrending structures
DimensionResult
Profitable windows / total5 / 8 = 62.5% (≥ 5/8 minimum threshold)
Full-history worst drawdown16.9% (was 26.1%)
Worst yearW1 2018 −14.4% (market −72%)
Best yearW3 2020 COVID +46.9%
Stage 1 (parameter stability)Robust near pyramid thresholds 5/12/60
Stage 2 (cross-regime consistency)5 regimes profitable
Stage 3 (bear resilience)2018 loss compressed from −30.6% to −14.4%

Conclusion: HonestTrend + pyramid + small position edge still exists, and drawdown control is better than the old config. W8 (2025-present) negative return needs continued observation — could be a new regime precursor.

  1. Live trade only BTC/USDT + ETH/USDT This is the clearest conclusion of this walk-forward. Adding more pairs = diluting edge. Set aside the “diversification” instinct, the data says what it says.

  2. Make regime filter a first-class citizen HonestTrend1mMTF already introduced 4h EMA as a gate. Suggested next version:

    • If 4h EMA 96 < 4h EMA 288 for > 7 consecutive days → pause new 1m/15m entries
    • Sacrifice a few rebound opportunities in W1 / W5, but avoid 2/3 of the drawdowns
  3. Keep kill-switch threshold at 20% W1 −30.6% is an annual number; broken down by month/quarter, the worst drawdown segment should be < 20%. The current risk_manager.py 20% threshold is reasonable.

  4. Watch W8 (2025-present) BTC+ETH only +2.31%, 4-pair version is still negative −3.85% — could be an early signal of market entering a new regime. Recommend rerunning this walk-forward monthly.

  5. Don’t build dedicated strategies for XRP/BNB These coins don’t fit pure technical trend strategies. If you want to trade them, you need to introduce on-chain/sentiment/event-driven signals (this project’s sentiment_snapshots and kol_events are built for that).

Terminal window
# 1. Pull historical data (~20 min)
scripts/download_bulk_binance.sh BTCUSDT 15m 2017-08
scripts/download_bulk_binance.sh ETHUSDT 15m 2017-08
# 2. Run walk-forward (~2 min)
python scripts/walk_forward_full_history.py \
--strategy HonestTrend15mDry \
--timeframe 15m \
--config configs/backtest/config_backtest_15m_btceth.json

JSON archived at walk_forward_history/full_history_HonestTrend15mDry_15m_<date>.json, usable for historical comparisons.