Paper Trading Lab

An automated trading system that runs three strategies on Alpaca paper-trading accounts, each funded with $1,000 of simulated money. The strategies trade on their own every market day, and this page rebuilds itself from the live accounts and the bots' decision logs.

Market closed 16 trades today16 trades since October 13 1 logged problem today Updated October 14 at 5:57 PM ET

QQQ breakout trader

$1,001.41

+0.14% since start+0.11% today

Staggered breakout portfolio

$1,018.67

+1.87% since start+1.13% today

Learning allocator

$1,012.23

+1.22% since start+0.49% today

S&P 500 (SPY), for comparison

$988.48

-1.15% since start-1.08% today

Today

Change in each account's value since the open, in five-minute steps.

Latest trades

Time (ET)	Trade	Amount
Oct 14, 3:57 PM	Sell XLPStaggered breakout portfolio	$338.99
Oct 14, 3:57 PM	Sell XLEStaggered breakout portfolio	$339.75
Oct 14, 3:57 PM	Sell SQQQStaggered breakout portfolio	$113.87
Oct 14, 3:57 PM	Sell SPXUStaggered breakout portfolio	$112.18
Oct 14, 3:57 PM	Sell SOXSStaggered breakout portfolio	$113.86
Oct 14, 3:57 PM	Sell SQQQQQQ breakout trader	$996.40
Oct 14, 3:40 PM	Buy SVXYLearning allocator	$198.63
Oct 14, 3:40 PM	Buy TQQQLearning allocator	$562.13

Since launch

The value of each $1,000 account since October 12, compared with putting $1,000 into the S&P 500 through SPY on the same day.

The strategies

Each strategy's chart shades the range that 80% of its backtested outcomes fell within after the same number of trading days. A result inside that range is ordinary luck; a result well outside it would suggest the strategy is no longer behaving the way it was tested.

QQQ breakout trader

Intraday · 3x leveraged

QQQ, the Nasdaq-100 fund, usually stays within a predictable distance of its opening price, and how far it typically strays depends on the time of day. Every 30 minutes the bot compares QQQ's price with the average move for that time over the past two weeks. A break above that range buys TQQQ, a fund that moves about three times as much as QQQ; a break below buys SQQQ, which moves three times in the opposite direction. The position is closed when the price falls back through the range or the day's volume-weighted average price, and everything is sold before the market closes.

Trades: TQQQ and SQQQ (3x long and short QQQ)
Decisions: Every 30 minutes, 10:00 AM to 3:30 PM
Based on: Zarattini, Aziz and Barbon (2024)
Backtest, unseen years: 38% a year from 2023 to 2026, worst drop 23%
Backtest, design years: 37% a year from 2017 to 2022, worst drop 27%

Holdings

No positions. The account is in cash.

Today's decisions

Time	QQQ	Normal range	VWAP	Position
13:30	728.94	734.81 – 746.82	733.82	Cash → SQQQ
14:00	728.03	734.82 – 746.81	733.18	SQQQ
14:30	728.67	734.94 – 746.68	732.74	SQQQ
15:00	727.75	734.90 – 746.73	732.50	SQQQ
15:30	729.07	734.85 – 746.77	732.04	SQQQ

Recent trades (2 in total)

Time (ET)	Side	Symbol	Amount	Price	Reason
Oct 14, 3:57 PM	Sell	SQQQ	$996.40	$36.08	end of day: never hold overnight
Oct 14, 1:30 PM	Buy	SQQQ	$995.30	$36.04	13:30 decision

Staggered breakout portfolio

Intraday · most active

This account applies the same breakout rule to three markets: the Nasdaq-100, the S&P 500 and the semiconductor index. Each market runs on three separate half-hour clocks offset by ten minutes, which splits the account into nine independent slices. Checking a single market more often than every 30 minutes made the strategy worse in testing, so staggering the clocks is how this account trades throughout the hour without giving up the edge.

Trades: TQQQ/SQQQ, UPRO/SPXU and SOXL/SOXS
Decisions: Every 10 minutes, 10:00 AM to 3:50 PM
Typical activity: About 16 trades a day
Backtest, unseen years: 21% a year from 2023 to 2026, worst drop 16%
Backtest, design years: 21% a year from 2017 to 2022, worst drop 25%

Holdings

No positions. The account is in cash.

Positions by clock

Market	:00/:30 clock	:10/:40 clock	:20/:50 clock
Nasdaq-100	SQQQ	Cash	SQQQ
S&P 500	Cash	Cash	SPXU
Semiconductors	Cash	Cash	SOXS

16 decisions and 4 position changes today. Each cell is one slice of the account: one market on one clock.

Latest decision (15:50, :20/:50 clock)

Market	Price	Normal range	VWAP	Position
Nasdaq-100	728.73	734.73 – 746.90	731.81	SQQQ
S&P 500	768.90	772.92 – 780.03	770.60	SPXU
Semiconductors	566.39	570.79 – 593.75	569.29	SOXS

Recent trades (9 in total)

Time (ET)	Side	Symbol	Amount	Price	Reason
Oct 14, 3:57 PM	Sell	XLP	$338.99	$86.02	end of day: never hold overnight
Oct 14, 3:57 PM	Sell	XLE	$339.75	$67.61	end of day: never hold overnight
Oct 14, 3:57 PM	Sell	SQQQ	$113.87	$36.07	end of day: never hold overnight
Oct 14, 3:57 PM	Sell	SPXU	$112.18	$34.08	end of day: never hold overnight
Oct 14, 3:57 PM	Sell	SOXS	$113.86	$42.41	end of day: never hold overnight
Oct 14, 1:30 PM	Buy	SQQQ	$1.08	$36.04	13:30 decision (:00/:30 clock)
Oct 14, 1:20 PM	Buy	SOXS	$111.98	$41.71	13:20 decision (:20/:50 clock)
Oct 14, 1:20 PM	Buy	SPXU	$111.98	$34.02	13:20 decision (:20/:50 clock)
Oct 14, 1:20 PM	Buy	SQQQ	$111.98	$35.81	13:20 decision (:20/:50 clock)

Learning allocator

Daily · adaptive

Five daily strategies share this account: Bitcoin trend-following, short volatility timed by the VIX futures curve, holding TQQQ overnight during uptrends, buying sharp two-day dips, and a 3x trend position sized to its recent volatility. An online-learning algorithm called Hedge decides how much money each one gets. It began with an equal split on October 13 and moves money toward the strategies that have been performing well, while keeping at least 5% in each so it can notice when one recovers.

Trades: TQQQ, SVXY and Bitcoin
Decisions: Daily at 3:40 PM; the overnight position is sold at the open
Backtest, unseen years: 23% a year from 2018 to 2026, worst drop 32%
Backtest, design years: 58% a year from 2012 to 2017, worst drop 16%

Holdings

Holding	Shares	Value	Gain
SVXY	3.0930	$199.22	$0.59 (+0.30%)
TQQQ	7.5896	$559.35	$-2.78 (-0.49%)

Current allocation (as of Oct 14, 3:40 PM)

Strategy	Holds	Signal	Share
Bitcoin trendHolds Bitcoin while its 28-day return is positive.	Bitcoin	In	20.9%
Short volatilityHolds SVXY while short-term volatility is below its three-month level.	SVXY	In	19.7%
Overnight TQQQHolds TQQQ from 3:40 PM to the next open while QQQ is in an uptrend.	TQQQ	In	19.3%
Dip buyingBuys TQQQ after a sharp two-day drop in an uptrend and sells on the rebound.	TQQQ	In	20.9%
3x trendHolds TQQQ while QQQ is above its 200-day average, sized to its volatility.	TQQQ	In (80%)	19.2%

How the allocation has changed

Each strategy's share of the account after each daily update. The gray line marks an equal 20% split.

Bitcoin trend

Short volatility

Overnight TQQQ

Dip buying

3x trend

Recent trades (5 in total)

Time (ET)	Side	Symbol	Amount	Price	Reason
Oct 14, 3:40 PM	Buy	SVXY	$198.63	$64.22	daily rebalance (learned (Hedge))
Oct 14, 3:40 PM	Buy	TQQQ	$562.13	$74.07	daily rebalance (learned (Hedge))
Oct 14, 3:40 PM	Sell	XLP	$338.12	$85.80	daily rebalance (learned (Hedge))
Oct 14, 3:40 PM	Sell	XLE	$339.70	$67.60	daily rebalance (learned (Hedge))
Oct 14, 3:40 PM	Sell	XLB	$331.59	$49.54	daily rebalance (learned (Hedge))

How it works

Research and testing

Each strategy began as published research or a well-known trading rule and was rebuilt and tested on this project's own data, with trading costs included. Design choices were made on one stretch of history and then checked once against later years the strategy had never seen. Every variant tested is logged, including the failures, and the bar for calling a result statistically significant rises with each attempt. 30 variants have been tried so far and none clears that bar, so the backtest figures here are evidence rather than proof.

Trading

The bots place market orders through Alpaca's paper-trading API, using fractional shares so a $1,000 account can hold leveraged funds and Bitcoin. Before going live, the intraday strategies were run through past trading days on a simulated clock to confirm that the live code makes the same decisions as the backtest, and it matched on more than 99% of them.

Infrastructure

The bots run on GitHub Actions and are started on schedule by cron-job.org. Every decision is written to a log that is committed to the repository. This page is rebuilt from those logs and the account data about every 30 minutes while the market is open, and the charts are drawn with Chart.js.