We tested 33 trading ideas. 27 didn’t make it.

6 with evidence27 rejected, not for CFDs, or analysis onlyS016S017S022S028D001D003

Each bead is one notebook from our research log. Move along the rail to read them.

ArchiBalt researches strategies for US index CFDs. Buy access to what’s left, with its rules, its live signals and every known weakness on the record.

Request access See the five strategies

Five strategies, two ideas.

Six notebooks had evidence, and they became these five. Four trade the rebound after a US sell-off and one trades the Fed calendar. The rebound versions often fire on the same days, so buying more of them adds less than the count suggests.

PRE-FOMC-US Paper trading nowLow trust Holds US indices into scheduled Fed statements, a drift documented since 2015. 2022 alone gave half the profit. Since 2023 it has earned +8.7 bp a trade (t 0.84). luck 2.09 needs 3.04 2.53 Per trade+19.9 bp Trades a year8 A year on notional+1.6 % Worst stretch−3.0 % REBOUND-US Paper trading nowLow trust Buys US indices in the session after a sharp US500 sell-off. Not statistically confirmed. Crisis years carry it: 40 signals in 2022, 8 in 2023. luck 1.52 needs 2.57 1.69 Per trade+31.8 bp Trades a year15 A year on notional+4.8 % Worst stretch−13.9 % REBOUND-US-CASH Paper trading now, in shadowMedium trust The same rebound, only when the fall happened inside the cash session. Its strong t was picked from four definitions after seeing the data. It still needs an independent check. luck 1.52 needs 2.59 3.15 Per trade+64.8 bp Trades a year11 A year on notional+7.2 % Worst stretch−7.2 % Rebound, early entry Ready to paper tradeLow trust The same rebound, entering at the very start of the next session. Not statistically confirmed. Without 2020 it earns about a quarter less per trade. luck 1.52 needs 2.57 1.65 Per trade+31.1 bp Trades a year15 A year on notional+4.7 % Worst stretch−15.4 % Rebound, cash-close signal Ready to paper tradeLow trust The same rebound, measuring the fall at the 16:00 New York close. The weakest of the five: t 1.08, and the median trade loses a little. luck 1.52 needs 2.57 1.08 Per trade+19.1 bp Trades a year16 A year on notional+3.1 % Worst stretch−17.2 %

t after costs what luck alone reaches what it needs, counting every variant we tried

Trust is the verdict of each strategy’s report card in the app. Low means important checks fail, so the backtest alone is not good evidence: paper or demo only. Medium means every backtest check passes, and it still has to be confirmed on new data.

Backtests on Capital.com hourly prices, May 2017 to October 2026, after spread and swap. All five curves use the same scale. 1 bp is 0.01 %. Figures are on the amount traded, before any leverage. Backtests are not future returns.

See every year, including the losing ones

Backtest result by year, in % of the amount traded, before any leverage
Strategy	2017	2018	2019	2020	2021	2022	2023	2024	2025	2026
PRE-FOMC-US	−0.0	+2.8	−0.2	+3.6	−1.5	+7.6	+1.3	+2.3	+0.6	−1.6
REBOUND-US	none	−2.9	+11.9	+14.3	+7.1	−4.3	+0.3	−2.7	+16.7	+5.2
REBOUND-US-CASH	none	+0.6	+4.1	+29.1	+4.5	+6.0	−0.4	+3.3	+20.2	+0.6
Rebound, early entry	none	−5.3	+10.5	+18.7	+8.0	−5.6	+0.0	−2.4	+14.4	+6.2
Rebound, cash-close signal	none	−8.9	+12.8	+12.0	+2.9	−9.6	−0.8	+0.6	+17.7	+2.1

All 5 strategies $10,000 a year $2,000 each

Request access

What you get for each one.

Your own ArchiBalt workspace, where each strategy you choose runs on live prices next to the evidence for it.

ArchiBalt never trades real money. If you act on a signal, you place that trade yourself, at your own broker.

The exact rules: Entry, exit, markets and sizing, in plain words. They are not on this page because they are what you are buying.
Live signals: The engine checks prices every hour and tells you by Telegram or email when a signal fires.
The full research: The report behind each strategy, with its charts, its tests and what it failed.
A report card: Whether to believe the backtest: neighbouring settings, higher costs, and how many tries it took.
A paper record you can check: Every signal and trade on live prices, with monthly statements and CSV exports.
Demo orders, if you want them: Send the trades to a Capital.com demo account, with limits and a kill switch. Virtual money only.

The REBOUND-US page in ArchiBalt: its trust rating, its evidence rail, its cumulative backtest and a table comparing research, backtest and paper results. — REBOUND-US in the app, from a test workspace with no paper trades yet. Its rules are hidden in this picture.

We make it hard for our own strategies to look good.

The bar rises with every try.

Test 31 ideas and one will look good by luck, so each variant we try raises the bar. PRE-FOMC-US reached t 2.53. As our 31st idea it needed 3.04, and it has not got there.

Failures stay on the record.

All 27 are still in the research log with their results. You can read every one of them on the rail at the top of this page.

Every number is after costs.

The spread of the entry candle comes off every trade, plus a night of swap where a position stays overnight. An edge that only exists before costs does not count.

The live engine has to match the report.

A strategy goes live only if the engine that trades it reproduces its research. One example: 151 events, +19.1 bp, t 1.08 in the report, and the same three numbers on the engine.

Read all 33 notebooks

Notebook	Outcome	Idea	Result
S001	Rejected	Close vs EMA12 on the first candle of the session, exit after 3 candles	12 061 trades, win rate 37.1 %, PF 0.51
S002	Rejected	S001 + strong first-candle filter	3 794 trades, PF 0.51
S003	Rejected	S002 + smaller back tail	1 594 trades, PF 0.52
S004	Rejected	S002 + no tail on the open side	231 trades, PF 0.42
S005	Rejected	S001–S004 on 15-minute candles	PF 0.37–0.58
S006	Rejected	S003 with EMA length 7–26	PF 0.49–0.55
S007	Rejected	S004 with EMA25	204 trades, PF 0.48
S008	Rejected	Asian-range breakout at the London open	3 515 trades, PF 0.97
S009	Not for CFDs	Intraday momentum: the last hour follows the day (Gao et al. 2018; Baltussen et al. 2021)	7 141 trades, +1.8 bp before spread, PF 0.99
S010	Not for CFDs	Overnight drift: long US500 02:00–03:00 ET (Boyarchenko et al. 2023)	2 347 trades, +0.54 bp before spread, PF 0.79
S011	Not for CFDs	Time-series momentum, 12 months (Moskowitz, Ooi, Pedersen 2012)	Sharpe 0.70 before costs, 0.32 with a 3 % swap
S012	Not for CFDs	Time-series momentum with real Capital.com swaps	Sharpe −0.08 after swaps (0.38 as futures)
S013	Rejected	Overnight drift after a US500 sell-off, tested on Europe	−2.4 bp per trade after spread
S014	Rejected	Rebound, daytime session only (09:00–16:00 ET)	+12.2 bp, t 1.49; without 2020 +1.8 bp; Europe −6.3 bp
S015	Rejected	S008 + volatility filter	out-of-sample −0.013 % vs −0.011 % without the filter
S016	Evidence	Rebound on daily data	+21.0 bp, PF 1.35; 2000–2016 +22.7 bp
S017	Evidence	Rebound on hourly candles	+31.9 bp, PF 1.52, event t 1.69
S018	Rejected	Same signal on 32 other markets	Non-US indices −16.9 bp and energy −31.5 bp before spread; only US2000 +17.2 bp after costs (t 1.69)
S019	Rejected	Logistic regression + dense neural net	Out-of-sample AUC 0.494 / 0.505
S020	Rejected	Short "catch-up" basket after a US500 sell-off	+18.4 bp per event (t 3.03), but 2017–2026 +9.2 bp (t 0.74)
S021	Rejected	Short energy after a US500 sell-off	−0.6 bp (t −0.02); +13.5 bp (t 0.36)
S022	Evidence	Rebound under three signal definitions	+16.3 / +31.7 / +16.8 bp per event, t 0.93 / 1.74 / 0.94
S023	Analysis only	US100 in the rebound: special market or beta?	US100 − US500 = +0.012 σ (t 0.36): mostly beta
S024	Rejected	ML filter (ridge, gradient boosting) for the rebound	Filter −1.5 bp vs always trading
S025	Rejected	Turn-of-the-month effect (McConnell, Xu 2008)	−3.2 bp after spread and swap
S026	Rejected	"Tech-led sell-off" condition for US100	Group difference −0.027 σ (t −0.38)
S027	Rejected	Rebound only in the top quintile of a ridge forecast	+55.4 vs +7.6 bp (t 1.37), no better than a plain filter
S028	Evidence	Drift before scheduled Fed statements (Lucca, Moench 2015)	+19.9 bp per event after costs (t 2.53); 2023–2026 +8.7 bp (t 0.84)
D001	Evidence	Rebound with a delayed entry	+31.8 bp, PF 1.53
D002	Rejected	Small transformer on daily bars	Validation −13.7 bp (PF 0.83), test +12.4 bp (t 1.0)
D003	Evidence	Rebound under four signal definitions	+31.8 / +19.1 / +21.4 / +64.8 bp per event, t 1.69 / 1.08 / 1.19 / 3.15
S029	Rejected	Transformer trained on simulated random orders, on Gold	+0.149 R per trade (t 2.09), but the same setups at random times give +0.104 R
S030	Rejected	Pre-Fed drift on European indices	+18.0 bp before costs, the same move as US500 that night; +7.1 bp after costs (t 0.86)

How it works.

Request access
Tick the strategies you want and send the form.
Go through the evidence
We reply by email and walk you through a full report on a call. Nothing to pay before that.
Get your workspace
Your strategies, their rules and their reports, ready and paper trading.
Decide on each signal
Signals reach you by Telegram or email. What you do with them is up to you.

Straight answers.

Will this make me money?

We can’t promise that. These are backtests and paper trades, and past results don’t predict future ones. CFDs are leveraged products that can lose money quickly.

Is any of this statistically proven?

No. Our own report card rates four of the five low trust: the backtest alone is not good evidence, so they are for paper or demo trading until new data confirms them. REBOUND-US-CASH rates medium. Every backtest check passes, but we picked it after looking at the data, and it still needs confirming on new data.

How much do I need to trade for this to be worth it?

Do this sum before you buy. In the backtest, REBOUND-US earned about 4.8 % a year on the amount traded and PRE-FOMC-US about 1.6 %, before leverage. Compare that with the price, and with a worst stretch of about 14 % for REBOUND-US.

Do you hold or trade my money?

No. ArchiBalt paper trades, and can send demo orders with virtual money. Any real trade is yours to place, at your own broker.

How often will I get a signal?

Rarely. Between 8 and 16 times a year per strategy, often in clusters, with months of silence in between. A strategy can go a long time without trading.

Which markets and which broker?

CFDs on US500, US100 and US30, tested on Capital.com’s hourly prices and spreads. Other brokers charge different spreads, which would change the results.

Can I see the research before paying?

Yes. We go through a full report with you on the call, before you pay anything.

Request access.

Tell us who you are and we’ll reply by email to arrange a call. There is nothing to pay until you’ve seen the evidence.

All 5 strategies

$10,000 a year

Indicative price for one year of access, confirmed on the call. Change your selection