DCA + Pyramid aggressive experiments

Experiment Report: Aggressive DCA and Trend-Strategy Pyramid Winners

Two independent enhancement experiments, both representing the correct form of “add to position on crash / add to position on profit”.

Experiment A: Aggressive DCA (FnG < 10 → 3× multiplier)

Hypothesis: The current DCA already uses a 2× multiplier when FnG < 20. Would being a bit more aggressive (FnG < 10 → 3×, cap 3.0) accumulate more BTC at a lower average cost?

Method

Time range: 2018-02-01 → 2026-04-17 (from the start of FnG data to the present)
Base amount: $500/week, buy every Monday
Cycle factor proxy: Mayer Multiple (price / 200-day MA) mapped to [-1, +1]
Script: scripts/backtest_dca.py

Comparison across 4 schemes

Scheme	Total Spent (USDT)	BTC Acquired	Avg Cost ($/BTC)	Portfolio Value (USDT)	PnL %	Extra vs Flat %
A. Flat $500/wk (baseline)	214,000	13.75	$15,569	$1,022,877	+378%	0%
B. Current formula (cycle 50% + FnG 30%)	328,200	22.00	$14,918	$1,637,210	+399%	+60.1%
C. Aggressive (FnG<10→3x, cap 3x)	356,812	24.52	$14,549	$1,825,060	+412%	+78.4%
D. FnG-only linear	250,461	17.48	$14,332	$1,300,496	+419%	+27.1%

Conclusions

Aggressive scheme (C) wins on absolute return: $1.82M, +78% over baseline
FnG-only (D) is the most efficient: +419% return per dollar, but lower absolute amount because it invests less
Current formula (B) is already decent: +60% vs flat. Upgrading to (C) adds another +18%
“Cost” of the aggressive scheme: $142,812 more spent (vs baseline), but $802,183 more earned → incremental ROI = 5.6×

Why is being aggressive effective?

FnG < 10 is extremely rare (only ~30 weeks across 8 years), so overcapitalization is not sustained long term
These moments are very likely cycle bottoms (2018-12, 2020-03 COVID, 2022-11 FTX)
Buying in extreme fear = buying near local bottoms → lower average cost

Recommendation

Adopt scheme C. Update compute_multiplier in strategies/dca_executor.py:

if fng < 10:
    fng_mult = 3.0          # NEW
elif fng < 20:
    fng_mult = 2.2          # was 2.0
# ...
multiplier = ... * kol_bonus
multiplier = max(0.0, min(3.0, multiplier))  # cap raised from 2.5 to 3.0

Visualizations

reports/dca_backtest/dca_comparison.html — 4-scheme portfolio comparison + cumulative investment + average cost
reports/dca_backtest/multiplier_distribution.html — multiplier frequency distribution per scheme

Experiment B: HonestTrend Pyramid Winners

Hypothesis: The trend strategy currently enters each position only once. If we add to the position after profit is confirmed (pyramid winners, not martingale losers), can total return be improved?

Method

New strategy: strategies/HonestTrend15mPyramid.py

Initial entry: EMA 94/139 golden cross + ADX > 18 + FnG < 80  (same as HonestTrend15mDry)
Pyramid 1: position profit ≥ +5%  → add 0.6× of initial size
Pyramid 2: position profit ≥ +12% → add 0.4× of initial size
Max total position: 2.0× initial (first 1.0 + 0.6 + 0.4)
Exit: EMA death cross (same as baseline)
Key rule: NEVER add to position on loss (this is not martingale)

Both runs use the same config (configs/backtest/config_backtest_15m_pyramid.json):

stake_amount: 3000 (fixed, leaves room for pyramid)
max_open_trades: 3
position_adjustment_enable: true (enabled for pyramid version)

Comparison results (2017-08-17 → 2026-04-20, BTC+ETH)

Metric	Baseline 15mDry	Pyramid 15mPyramid	Difference
Trades	570	570	0
Win Rate	38.2%	33.2%	−5.0 ppt
Total Profit (USDT)	$26,823	$37,782	+$10,959 (+41%)
Total Profit %	+268%	+378%	+110 ppt
Avg Profit / trade	1.20%	0.15%	−1.05 ppt (because position is larger)
Max Drawdown	25.50%	24.03%	−1.5 ppt

Why does Pyramid work?

Winners get bigger: an average $1,439 winner gets scaled to an effective $2,000+ (initial $3K + 0.6× $3K added at +5% profit, then +0.4× at +12%)
Losers do not get bigger (rule: only add on profit)
Win rate actually drops (−5 ppt): some trades that were at +5–12% entered pyramid and then fell back below 0, counted as losers
But average win × number of wins > average loss × number of losses still holds → net return increases

Why does Max DD shrink?

Counter-intuitive but supported by the data:

Pyramid is a “profit-contingent” add-on, mostly occurring during bull/rising phases
Bull phases push the equity peak higher, so subsequent bear drawdowns start from a “higher base” — in percentage terms they are smaller
And because we never add on losers, the worst-case downside is not amplified

Adopted (2026-04-21) — Mode 2 aggressive integration

adjust_trade_position and position_adjustment_enable = True have been added directly to the HonestTrendGeneric base class. All subclasses inherit automatically:

HonestTrend15mDry — active
HonestTrend1mLive — active
HonestTrend1mMTF — active

The 3 active configs have been updated:

configs/config_dryrun_honest15m.json
configs/config_dryrun_honest1mmtf.json
configs/config_live_honest1m.json

Each config adds:

"position_adjustment_enable": true,
"max_entry_position_adjustment": 2,
"stake_amount": 1500   // was "unlimited"

stake_amount changed from “unlimited” to a fixed 1500 because:

The original unlimited value would fill the entire wallet (3 trades × 3333 = 9999), leaving no room for pyramid
The new value of 1500 leaves room for pyramid (a single trade can grow to 1500 + 900 + 600 = 3000; 3 parallel trades use at most 9000, with a 1000 buffer)
Before going live, be sure to scale this to your actual capital allocation for HonestTrend (stake = dedicated_capital × 0.15)

The experimental files have been deleted:

~~strategies/HonestTrend15mPyramid.py~~ (merged into the base class)
~~configs/backtest/config_backtest_15m_pyramid.json~~ (equivalent to btceth)

Visualizations

reports/pyramid/index.html — full 7-panel dashboard
Compare with reports/full_history_btceth/index.html (baseline)

“Philosophical” distinction between the two experiments

	Aggressive DCA	Pyramid Winners
When to add	On crash / panic	On confirmed profit
Trigger condition	FnG / cycle undervaluation signal	Current position > +5% / +12%
Nature	Contrarian trading (buy more as it drops)	Trend-following addition (buy more as it rises)
Suitable for	Long-term accumulation (spot holdings)	Active trend strategy
Implementation location	`dca_executor.py` weekly timer	`HonestTrend15mPyramid.adjust_trade_position`

The two can coexist without conflict because they act in different market phases:

DCA adds at well-defined bear bottoms → buy for long-term accumulation
Pyramid adds within a well-defined bullish trend → amplify the current trade

Never mix them into “add on loss” (Martingale) — that would destroy the benefits of both.

Reproduction commands

# DCA experiment
python scripts/backtest_dca.py
# → reports/dca_backtest/summary.csv + *.html

# Pyramid experiment
freqtrade backtesting --strategy HonestTrend15mDry \
  --config configs/backtest/config_backtest_15m_pyramid.json \
  --datadir user_data/data/binance \
  --user-data-dir user_data \
  --strategy-path strategies \
  --timerange 20170817- --export signals

freqtrade backtesting --strategy HonestTrend15mPyramid \
  --config configs/backtest/config_backtest_15m_pyramid.json \
  --datadir user_data/data/binance \
  --user-data-dir user_data \
  --strategy-path strategies \
  --timerange 20170817- --export signals

python scripts/visualize_strategy.py --out reports/pyramid