---
name: ma-stack-study
description: Backtest (4 Oct 2026) of Ben's MA100/MA200 "flat stacked base, break, retest, hold" setup on 1h, 4h and daily, plus higher-timeframe filters, daily MA100 support/resistance and what to wait for after a pump. Read before any MA-based entry call.
sources: [cowork]
aliases: [ma strategy, flat ma setup, stacked mas, ma100 ma200, daily ma100, ma retest]
---
- [stated] Ben's thesis (4 Oct 2026): look for coins where the 1h/4h/daily MA100 and MA200 go flat and sit on top of each other (accumulation), price closes above them, retests them, holds, then lifts off; avoid times when the MAs are far apart and oscillating; prefer no higher-timeframe MA directly overhead; daily MA100 acts as support/resistance on a closing basis; young coins use lower timeframes first. Examples he gave: PONS 4h (Aug 2026), UTYA daily, AKE, UAI, WOJAK.
- [stated] He also said MAs are zones, not exact lines: price gets front-run or dumped through them.

Method: GMGN token_kline (droplet key) 1h/4h/1d, 732 tokens (RH, Solana, BSC universe of ~200 per chain + old Sol/BSC coins), closing basis, filters: 24h volume >= $10k, typical daily move >= 2.5% (drops stables, LSTs, stock tokens). Entry after break + retest (zone: low within 5% of the stack, close >= 97% of lower MA) + hold (6 bars 1h, 3 bars 4h/1d). Outcome: +50%/-20% close-basis bracket, 7d (1h) / 14d (4h) / 30d (1d), 2% cost, minus the same coin's random-bar result (xR, pts). Controls: same setup on lines shifted +-12/20%, tight-range breakout with no MA, coin-clustered bootstrap 90% CI, time halves. GeckoTerminal public API caps history at 180 days (no daily MA200 there); GMGN gives full life. GMGN closes match GT within ~1%.

Results (219 coins 1h, 200 4h, 185 daily; RH only ~60 coins pass, too young/thin for daily):
- Setup alone: no edge on any timeframe. 1h xR -0.5 [-1.8,+0.6], 4h +0.9 [-0.7,+2.5], daily +0.7 [-3.3,+6.1]. Fake shifted lines do about the same, so the exact MA level is not special.
- Choppy (MAs far apart) setups lose: 4h -7.3 [-12.2,-2.7], daily -4.2 [-6.8,-1.0], both halves negative. Ben's "skip when MAs oscillate" holds. PROVEN.
- 1h setup while price is above the 4h MAs but the 4h MAs are apart: -5.5 [-7.9,-3.3] (at $50k volume -4.0 [-6.2,-1.3], both halves). Skip these. PROVEN.
- 1h setup with 4h MAs stacked and price above them: +2.0 [-0.3,+4.8], new ATH later 61% vs 36% base. Weak plus. TESTING.
- 4h setup with daily MAs stacked and price above: +6.9 [-1.2,+15.3], n=64, 17 coins; at $50k volume +8.0 [+1.4,+17.7], halves +5/+9, new ATH 55%. Best combo, small sample. TESTING.
- "Gentle up" MA200 slope bonus: not supported (4h -0.8, daily -6.4). Flat or down slope did slightly better on 4h (+2.6 [+0.1,+5.2]).
- MA200 just formed (Pons type): 1h -6.8 [-10.9,-2.7]; 4h +2.7 wide. Not general.
- Young coins (first third of data) on 1h: -5.7 [-10.7,-0.3]. The "young coins follow 1h" idea not supported on 1h; 1m to 15m not tested yet.
- Daily MA100: retest after a reclaim +3.0 [-0.6,+6.7] vs fake lines -0.4 (TESTING); wick-rejection from below -0.7 vs fakes +0.8 (weak resistance); after a break, share of next 10 days closing above 58% vs 59% on fake lines (not special). Simply being above daily MA100 was slightly worse next 30 days (-1.8 [-2.7,-0.8]), so it is not a blanket filter.
- After a 3x+ pump (198 pumps): random later bar makes a new high 30%; first 1h setup after the top 35% vs 31% same-pump base (n=81), forms ~12 days after the top at ~-50%; 4h setups rarely form within 30 days (18 of 198). If the 4h MAs are >30% above price, new high after a 1h setup 0 of 21. Riding the 1h MA100 on the way up did not make the later 1h setup better.
- PONS receipts: 1h setup 20 to 22 Aug 2026 (4h MAs just formed, price above) ran +237% to +315% in 7 days; ten 1h "flat stacked" setups 17 to 27 Sep (after the pump, 4h MAs apart) made 0 to 20%. On 4 Oct PONS 4h: price -30% under MA100, -33% under MA200, MA gap 4%, MA100 falling 6% per week.
## v2, 4 Oct 2026 (Ben: more coins, smaller timeframes, all-aligned test, winner patterns)
- [stated] Ben's v2 asks: scale up until reliable; test 5m/15m on young Robinhood coins; theory that BSC is more technical; theory that the best setup is every timeframe aligned incl daily (break above daily, retest, go); stay away while 4h MAs chop; find what 100%+ winners after 15m/1h/4h flips share vs losers.
- Data: 2,728 tokens (GMGN trending ranks x 3 chains x many sorts/age windows + GeckoTerminal lists), GMGN 5m/15m/1h/4h/1d (15m = last 10 days, 5m = last 3.5 days for coins under 16 days old). Usable after filters: 5m 272 coins, 15m 1,152, 1h 934, 4h 503, daily 236 (BSC/Sol only, RH too young for daily MA200). Extra exit tested: "let it run" = hold to long horizon, stop on a close 30% under entry, capped at +900%.
- PROVEN: next-higher timeframe above but its MAs apart (chop) kills the flip: 15m -4.5 [-5.6,-3.4], 1h -5.9 [-7.4,-4.4], 5m -5.2; both halves, all chains.
- PROVEN: choppy setups lose on every timeframe (5m -5.4, 15m -4.5, 1h -5.5, 4h -4.9, 1d -4.7).
- PROVEN: flips after a coin already ran 4x+ in 30 days lose: 15m -4.7, 1h -4.9, 4h -3.9 (let-run -10 to -13).
- 4h flat stacked setup is the only timeframe with a plus: +3.0 [+1.6,+4.5], let-run +8.0 [+3.1,+13.8]; BSC let-run +10.4 [+4.3,+17.8]. First half of each coin's data was negative (-2.7), so TESTING not PROVEN.
- 15m and 5m flips lose after costs (15m flat -1.0, 5m -3.1); young coins (0 to 7 days) on 5m/15m no better. Small timeframes: KILLED as a standalone entry.
- All aligned incl daily: 4h setup +5.2 [+0.3,+10.9] (n=149, 31 coins) on the bracket exit but let-run only +1.7 (n.s.); 1h all aligned +1.5 (n.s.). Helps a bit, not the big-run key. TESTING.
- Daily MAs overhead within 30% did NOT hurt 4h flips (bracket +2.5, let-run +10.4). The "daily is resistance overhead" filter is not supported.
- BSC more technical: partly yes. BSC is the only chain where 1h flips are flat to positive (+0.3 vs RH -4.8, Sol -1.4), best 4h runs, and chop is punished hardest on BSC 15m (-11.3). Robinhood is worst on 15m/1h.
- Winners (+100%) vs losers: winners are young (median 9 days on 15m, 36 on 1h), tiny cap (median $160k on 15m, $890k on 1h), already ran ~7x, deep under ATH, wide range, MAs apart, short or no flat base. Out of sample a model finds them (AUC 0.83 15m, 0.74 1h, 0.62 4h) but those traits also make big losers, so the average trade is not better except tiny young coins on 1h (under 27 days and under $420k: let-run +25 [0,+58], n=227, 89 coins). KILLED same day on drill-down: it was every 1h flip type (147 of 227 were choppy), 68% hit the -30% stop, median -34%, five coins (UBI Lyi47medADEVDd5hxJo1mbxhnBct841sFpcGRyHTuwp, TIMELESS 0x25b865486dac88b811d8600494a1f99387bb7777, ALICE 2hkhFLWQKUcvUNh7eUkfZVz3rYfxXqFNeVLYMotrzFky, PURR 8RNUw4N655VSrZKuhGdywhbSMDTrheguFPfxbpE2NZHQ, BABYCATE DDVUsN8sDFxbaX6gNBoD44kjZhFETWJnwAn4EX1dpump) made all the profit, the rest averaged 0; market cap was only known for coins on today's GMGN trending list (survivor bias). Best expectancy on 1h came from the opposite profile: tight base, near ATH, long flat base (let-run +1 to +3, n.s.).
## v3, 4 Oct 2026: out-of-sample run on NEW coins (Ben: 300 coins never looked at, Solana or Robinhood, typical gains, test the skip rules)
- Sample: 861 new tokens (Jupiter verified Solana list, first pool 60+ days old, random pick incl. now-dead coins, plus 9 new RH from GMGN); 427 usable on 4h (425 Solana, 2 RH). Robinhood has almost no unseen coins old enough for a 4h test (chain too young; the old-enough ones were all in v2). GMGN key got an escalating 429 ban when run with 14 threads; 4 threads at 0.6s gap ran clean.
- These coins fell hard in the window: random entry let-run median -32%, average -13%, 20% ended up.
- Main 4h flat stacked setup: 318 trades on 136 coins. Let-run (hold 30d, sell on close -30%): median -18%, average 0%, 25% ended up, 41% stopped, 10% made +50%, 4% doubled. Quick exit (+50%/-20%, 14d): median -13%, average -1.5%.
- Edge vs random entry in same coins: +7.8 pts [-3.1,+20.1], halves +4.7/+11.7. Same size as v2 (+8.0), but not significant alone. Fake lines +4.3.
- Typical gains depend on the coin set: v2 sample (rising coins) median +1%, average +16%, 51% ended up; new sample (falling coins) median -18%. The setup beats random by ~8 pts in both, it does not turn a falling market green.
- Skip rules: daily-chop skip kept +10.3 vs skipped -1.6 (same direction as v2 +15.2 vs -12.0, n.s. here). The 4x-prior-run skip almost never fires on 4h flat setups (4 trades). All-aligned (daily stacked, price above): 7 trades, untestable.
- Both samples: 4h flips with the daily MAs OVERHEAD did better than flips above the daily MAs (v2 overhead_far +16 pts, overhead_near +12, daily above -5 to -12; new overhead_far +17, overhead_near +4, above_loose -2). So "everything aligned" is NOT supported; the better flip is a 4h base breaking out under the daily MAs. TESTING.
- Raw: Mac ~/ma/gm3b.tgz (new coins), s3_h4.pkl in ~/ma results; script new4h.py.
- Raw: Mac ~/ma/gm2.tgz (2,728 token candles), scripts setup2.py, an3.py, win.py, rule.py in ~/ma/ma_study_v2.tgz.
- Raw: Mac ~/ma/gm_all.tgz (candles) and ~/ma/ma_study_results.tgz (scripts setup.py, an2.py, combo.py, daily.py, pump.py, results); droplet /root/gm.