MemLabs Trading Competition #1

⚡ ACTIVE BETA

Aug 17, 2026 to Dec 31, 2026 102 days left 13 participants

1 day bars Metric: Average Trade Return

Prize Pool for Top 3 Submissions

Introduction to Quant Recruiters for top 3 submissions
Free Annual MemLabs Subscription worth over $3200
MemLabs Competition Winner T-Shirt for top 5 submissions
Winner is added to the Hall of Fame

Overview

A competition to test out your quant skills. Your task is to build a model to predict the future daily return. The model that makes the most money wins. You are given a time series of daily prices and must predict tomorrow's return.

The data is split chronologically into two sets:

1 Training data: Earlier observations used to develop and refine your model.
2 Test data: The most recent observations, held out from model development. This out-of-sample period measures whether your model generalizes to unseen data and remains robust when market behaviour changes.

Goal

Maximize the average trade return:

\max \; \frac{1}{N}\sum_{t=1}^{N} R_t^{\mathrm{trade}}

Definitions

$R_t^{\mathrm{trade}}$ : trade return at time t
$N$ : number of evaluated predictions
$s_t \in \{-1, 0, 1\}$ : position at time t: short, flat, or long
$\hat{r}_{t+1}$ : model's prediction of tomorrow's return
$r_{t+1}$ : tomorrow's realized return

The evaluator computes:

R_t^{\mathrm{trade}} = s_t \times r_{t+1}

Each prediction is converted to a directional signal.

s_t = \operatorname{sign}(\hat{r}_{t+1}) = \begin{cases} 1, & \hat{r}_{t+1} > 0 \\ 0, & \hat{r}_{t+1} = 0 \\ -1, & \hat{r}_{t+1} < 0 \end{cases}

-1 = your model predicts the return goes down
1 = your model predicts the return goes up

Example

Model predicts the next return is 0.1% -> signal = 1
Model predicts the next return is -0.2% -> signal = -1

Rules

1 Use only the provided competition dataset. No external data sources or APIs are permitted, but you may create features (model input) from the competition dataset.
2 Your model must predict the future daily return - tomorrow's return.
3 Return a numeric forecast. The evaluator converts it to a signal with np.sign(y_hat).
4 Predictions are converted with np.sign(y_hat): down is -1, zero is 0, and up is 1.
5 The evaluator computes trade_return = signal * actual_return, where actual_return is the next close-to-close simple return.
6 The leaderboard score is the average trade return on the unseen test data. The average trade return must be positive on both the training and test data. Values extremely close to zero are treated as zero.
7 Submissions found to be cheating will be disqualified
8 The following Python libraries are allowed: NumPy, scikit-learn, XGBoost, LightGBM, Keras, JAX, and jaxlib.
9 The live leaderboard is provisional. When the competition closes, additional unseen data will be added to the test dataset and eligible submissions will be evaluated again. This final evaluation will determine the official rankings and winners.
10 Each participant may complete one evaluation per UTC calendar day, based on when it was submitted. Failed evaluations can be retried that day. Completed evaluations count regardless of score or rule violations. Only one evaluation may be queued or scoring at a time.

Scoring

How your leaderboard score is calculated

1. Evaluate the strategy on the training and test datasets separately.
2. Calculate the average trade return for each dataset.
3. Use the average from the unseen test data as the leaderboard score.

\mathrm{Score} = \overline{R}_{\mathrm{test}}

To qualify, the average trade return must be positive on both datasets:

\overline{R}_{\mathrm{train}} > 0 \quad \text{and} \quad \overline{R}_{\mathrm{test}} > 0

Values extremely close to zero are treated as zero to avoid rounding errors.

Average Trade Return (Primary)

Your average trade return on the unseen test data. Positive training and test averages are required to qualify.

Leaderboard Entry Criteria

Positive train and test returns

Secondary Metrics

Win Rate

Percentage of active trades with a positive trade return.

Ann. Sharpe Ratio

Annualised return divided by annualised volatility. Higher is better; above 1 is considered good.

Max Drawdown

The largest percentage decline from a previous peak in the compounded equity curve.

Best Trade

The single highest trade return achieved during the evaluation period.

Worst Trade

The single lowest trade return during the evaluation period.

Live Leaderboard

13 participants

	User	Model
1	Peter Heinum	dead_cat_ladder	+0.0062	0.0332	3.55	61.0%	-0.1706	+0.1499	-0.1163	Private
2	Jahy	linear3	+0.0059	0.0333	3.40	56.3%	-0.2529	+0.1499	-0.1163	Private
3	zenji too	quant_zenji_instagram2	+0.0051	0.0333	2.95	56.7%	-0.2282	+0.1499	-0.1163	Private
4	144p Yasuo	momentum	+0.0038	0.0336	2.14	53.0%	-0.3816	+0.1163	-0.1499	Private
5	fuji	WinniethePoo	+0.0035	0.0319	2.12	54.7%	-0.2124	+0.1499	-0.1163	Private
6	arjen9	ExMachina	+0.0029	0.0336	1.67	56.0%	-0.2738	+0.1499	-0.1163	Private
7	Billy	Exponential Fair Value	+0.0021	0.0230	1.77	55.1%	-0.2935	+0.1084	-0.1161	Private
8	Balaenys.	i suck	+0.0019	0.0337	1.09	50.7%	-0.3129	+0.1499	-0.1163	Public
9	memlabs	Simple Momentum Model	+0.0019	0.0337	1.09	50.7%	-0.3129	+0.1499	-0.1163	Public
10	randseed_trader	2	+0.0016	0.0167	1.80	56.2%	-0.1271	+0.1161	-0.1153	Private
11	Lars van Oostrum	gg2	+0.0013	0.0276	0.91	51.2%	-0.3351	+0.1163	-0.1499	Private
12	householddude	?	+0.0010	0.0338	0.55	53.0%	-0.3546	+0.1499	-0.1163	Private
13	yoitspeter	xgb_save_me	+0.0003	0.0338	0.17	48.0%	-0.5793	+0.1163	-0.1499	Private

Training Data

Anonymised Asset · 1 day bars