Much of the variation in financial time series is driven by market-wide shocks that cannot be predicted from an asset's own history. Per-asset self-supervised learning treats this variation as irreducible noise. Under a latent factor model, we show that conditioning prediction on the realized market factor makes the asset's market exposure necessary for optimal prediction, while reducing the noise against which the asset-specific component is learned. Building on this insight, we introduce Market-Conditioned JEPA (MC-JEPA), a joint-embedding predictive architecture for financial time series. During training, we construct a leave-one-out market condition for each asset from the future target embeddings of other assets and incorporate it through factor loadings derived from the asset's own representation. On 11 years of minute-level Chinese A-share data, MC-JEPA improves over per-asset JEPA across nine targets spanning returns, market exposure, risk, and liquidity, with the strongest and most statistically significant gains in exposure, risk, and liquidity. It achieves the best RankIC on all liquidity targets and on return horizons from two hours to five days, outperforming time-series foundation models and self-supervised baselines, and delivers the best portfolio performance in a historical backtest. Ablations show that the return and exposure gains require the condition to be aligned with the realized market movement. After restoring the volatility scale removed by normalization, the predictor's factor loadings rank stocks by market beta nearly as well as a supervised probe, without labels. A synthetic factor-model study with known parameters shows that conditioning improves the recovery of both market exposure and the asset-specific component.
Figures & tables
Figure 1: MC-JEPA. During training, each asset’s future embedding is predicted both unconditionally and conditioned on a market factor derived from the future target embeddings of other assets. The factor enters through loadings computed from the asset’s own representation. At inference, only the per-asset context encoder fθ is retained.
Return
Exposure
Risk
Liquidity
Method
Ret 120m
Ret 1d
Ret 5d
Ret 20d
Beta 20d
Vol 120m
Volume 120m
Volume 1d
Open vol share
Alpha158
.0425
.0591
.0867
.1011
.356
.364
.257
.417
.353
Kronos-mini
.0521
.0652
.1028
.1395
.324
.397
.466
.367
.358
Kronos-small
.0601
.0673
.1035
.1418
.333
.419
.552
.423
.376
Kronos-base
.0647
.0649
.1048
.1409
.348
.450
.568
.433
.381
TimesFM-2.5
.0564
.0626
.0983
.1358
.350
.424
.515
.406
.373
Table 1: Main results on the 2023–2025 test set. We report RankIC of ridge probes fitted to frozen encoder representations or Alpha158 features. Encoders trained in our experiments report mean ±std over three seeds. Bold and underlined values indicate the best and second-best scores, respectively.
Method
AER ↑
Sharpe ↑
IR ↑
MDD ↓
Alpha158
0.103
1.07
1.16
0.135
Kronos-mini
0.128
1.09
1.32
0.159
Kronos-small
0.101
0.95
1.12
0.177
Kronos-base
0.107
0.99
1.16
0.161
TimesFM-2.5
0.096
0.89
1.07
0.190
Chronos-2
0.092
0.86
1.01
0.186
Table 2: Backtest results on the 2023–2025 test period. Each day, the strategy selects the 7–8 stocks with the highest ridge-predicted 20-day returns among point-in-time CSI300 index members and holds them for 20 trading days. Results are net of 5/15 bp transaction costs, with excess returns measured against the CSI300 index. AER denotes annualized excess return, Sharpe the annualized Sharpe ratio, IR the annualized information ratio of the excess return, and MDD the magnitude of maximum drawdown. Encoders trained in our experiments report mean ±std over three seeds. Bold and underlined values indicate the best and second-best scores, respectively.
Return
Exposure
Risk
Liquidity
Variant
Ret 120m
Ret 1d
Ret 5d
Ret 20d
Beta 20d
Vol 120m
Volume 120m
Volume 1d
Open vol share
(a) Components
Per-asset JEPA
.0677 ±.0026
.0712 ±.0009
.1038 ±.0006
.1337 ±.0025
.359 ±.003
.444 ±.017
.573 ±.005
.440 ±.003
.391 ±.002
Panel JEPA ( λ=0 )
.0663 ±.0013
.0714 ±.0004
.1042 ±.0010
.1354 ±.0006
.345 ±.005
.439 ±.029
.572 ±.004
.439 ±.003
.388 ±.001
Placebo condition (other date)
.0667 ±.0044
.0720 ±.0015
.1031 ±.0016
.1350 ±.0036
.362 ±.002
.487 ±.023
.584 ±.003
.448 ±.004
.392 ±.003
Unfactored predictor
.0658 ±.0006
.0732 ±.0015
.1066 ±.0011
.1365 ±.0013
.366 ±.002
.498 ±.016
.589 ±.006
.450 ±.004
.394 ±.001
Table 3: Ablation results on the 2023–2025 test period. We report RankIC of ridge probes fitted to frozen representations, using the targets in Table 1 . Each variant changes only the indicated component or hyperparameter relative to MC-JEPA. Results are reported as mean ±std over three seeds. Bold values indicate the best score in each column.
Figure 2: Synthetic factor-model results. We report the R2 of linear probes fitted to frozen representations for market exposure Bi (left) and the asset-specific component ai (right), evaluated on held-out assets. The x-axis varies the predictability of the next market factor, measured by the R2 of predicting ft+1 from its history. Points show the mean and error bars the standard deviation over five seeds.
Appendix figures & tables13 assets
Supplementary material from the paper’s appendix.
Appendix
Setting
Value
Patch length p
10 bars
Transformer layers
12
Width D / MLP width
256 / 1024
Attention heads
8
Encoder parameters
9.52M
Horizons H
{1,60,240} bars
Appendix
Table 4: Encoder and predictor config.
Setting
Value
Decision times per step
128
Condition pool ∣Mt∣
32
Scored assets ∣St∣
8
Condition weight λ
0.75
Optimizer
AdamW, β=(0.9,0.95)
Learning rate
3×10−4
Appendix
Table 5: Training settings of MC-JEPA.
Model
Checkpoint
Kronos-mini
NeoQuasar/Kronos-mini
Kronos-small
NeoQuasar/Kronos-small
Kronos-base
NeoQuasar/Kronos-base
TimesFM-2.5
google/timesfm-2.5-200m-pytorch
Chronos-2
amazon/chronos-2
MOMENT-small
AutonLab/MOMENT-1-small
Appendix
Table 6: Checkpoints of the pretrained time-series foundation model baselines.
Figure 3: Cumulative excess return over the CSI300 index of the backtest in Table 2 , net of transaction costs. For the encoders trained in our experiments, lines show the mean and shaded bands the range over three seeds.
Figure 4: Share of the unconditional error removed by the market condition at each horizon h , 1−ℓhc/ℓh0 , on the validation period at the end of training, (a) by number of factors K at ∣Mt∣=32 and (b) by condition pool size ∣Mt∣ at K=1 . Points show the mean and error bars the standard deviation over three seeds. The circled point is MC-JEPA in both panels.
Return
Risk
Liquidity
Method
Ret 5m
Ret 30m
Ret 60m
Vol 5m
Vol 30m
Vol 60m
Volume 5m
Volume 30m
Volume 60m
Alpha158
.0459
.0629
.0607
.256
.347
.359
.172
.195
.223
Kronos-mini
.0546
.0682
.0625
.146
.339
.371
.292
.358
.407
Kronos-small
.0600
.0683
.0637
.188
.364
.395
.366
.441
.489
Kronos-base
.0603
.0686
.0634
.207
.394
.425
.381
.455
.502
TimesFM-2.5
.0599
.0727
.0696
.159
.352
.387
.343
.410
.457
Appendix
Table 7: RankIC of the minute targets at horizons of 5, 30 and 60 minutes, decided at the 11:30 morning close. The 120-minute horizon is in Table 1 . Encoders trained in our experiments report mean ±std over three seeds. Bold and underlined values indicate the best and second-best scores, respectively.
Risk
Liquidity
Method
Vol 1d
Vol 5d
MDD 5d
Volume 5d
Alpha158
.582
.562
.590
.495
Kronos-mini
.540
.522
.492
.451
Kronos-small
.540
.518
.505
.496
Kronos-base
.551
.532
.507
.504
TimesFM-2.5
.551
.537
.520
.486
Appendix
Table 8: RankIC of additional daily targets, decided at the 15:00 close. Encoders trained in our experiments report mean ±std over three seeds. Bold and underlined values indicate the best and second-best scores, respectively.
Return
Risk
Liquidity
Variant
Ret 5m
Ret 30m
Ret 60m
Vol 5m
Vol 30m
Vol 60m
Volume 5m
Volume 30m
Volume 60m
(a) Components
Per-asset JEPA
.0730 ±.0025
.0767 ±.0036
.0716 ±.0054
.240 ±.011
.408 ±.017
.430 ±.017
.424 ±.002
.481 ±.003
.517 ±.004
Panel JEPA ( λ=0 )
.0718 ±.0008
.0758 ±.0024
.0717 ±.0036
.243 ±.006
.408 ±.023
.426 ±.027
.425 ±.003
.483 ±.004
.518 ±.004
Placebo condition (other date)
.0729 ±.0017
.0738 ±.0025
.0716 ±.0065
.262 ±.010
.447 ±.019
.470 ±.021
.432 ±.005
.491 ±.004
.528 ±.003
Unfactored predictor
.0707 ±.0017
.0730 ±.0013
.0695 ±.0058
.270 ±.004
.462 ±.012
.483 ±.013
.437 ±.004
.497 ±.005
.534 ±.005
Appendix
Table 9: Ablation results on the minute targets of Table 7 . Variants and seeds are as in Table 3 . Bold values indicate the best score in each column.
Risk
Liquidity
Variant
Vol 1d
Vol 5d
MDD 5d
Volume 5d
(a) Components
Per-asset JEPA
.512 ±.007
.495 ±.007
.516 ±.000
.513 ±.003
Panel JEPA ( λ=0 )
.485 ±.015
.467 ±.016
.510 ±.004
.513 ±.002
Placebo condition (other date)
.517 ±.012
.499 ±.011
.518 ±.003
.522 ±.005
Unfactored predictor
.527 ±.004
.507 ±.004
.520 ±.001
.524 ±.005
Appendix
Table 10: Ablation results on the daily targets of Table 8 . Variants and seeds are as in Table 3 . Bold values indicate the best score in each column.
Return
Exposure
Risk
Liquidity
Variant
Ret 120m
Ret 1d
Ret 5d
Beta 20d
Vol 120m
Volume 120m
Volume 1d
Open vol share
(a) Condition weight λ
λ=0.25
.0620 ±.0043
.0707 ±.0021
.1146 ±.0013
.311 ±.001
.499 ±.002
.614 ±.002
.474 ±.003
.405 ±.001
λ=0.5
.0671 ±.0031
.0713 ±.0035
.1159 ±.0051
.312 ±.002
.514 ±.008
.614 ±.007
.474 ±.002
.405 ±.001
λ=1
.0649 ±.0032
.0732 ±.0022
.1171 ±.0023
.319 ±.003
.517 ±.009
.619 ±.004
.477 ±.001
.406 ±.002
(b) Number of factors K
Appendix
Table 11: Validation results (2022-07 to 2022-12) for the condition weight, factor-count and condition pool variants of Table 3 , on its targets. Probes are fitted on 2019-01 to 2022-06, as in Table 3 . Seeds are as in Table 3 . Bold values indicate the best score in each column.
vs. Per-asset JEPA
vs. Placebo condition
Target
Δ
t
Seeds
Δ
t
Seeds
Return
Ret 5m
0.0011
1.3
2/3
0.0013
1.8
2/3
Ret 30m
− 0.0013
− 1.1
1/3
0.0016
1.7
1/3
Ret 60m
0.0000
0.0
1/3
0.0000
0.0
1/3
Ret 120m
0.0018
1.5
2/3
0.0028
3.2
2/3
Appendix
Table 12: Paired tests of MC-JEPA against per-asset JEPA and the placebo condition on the 2023–2025 test set. Δ is the mean daily RankIC difference, and t is its Newey–West t -statistic, with 20 lags for Ret 20d and Beta 20d and 6 otherwise. Bold marks t>1.96 , a significant improvement at the 5% level. Seeds is the number of seeds, out of three, for which MC-JEPA has the higher RankIC.
Features
Beta 20d
MC-JEPA
Representation zi
0.374 ±0.003
Loadings bψ(zi,240)
0.366 ±0.003
zi + loadings
0.377 ±0.002
Trailing betas
0.525
MC-JEPA zi + trailing betas
0.539 ±0.001
Appendix
Table 13: Ridge probes for Beta 20d on the test period. Trailing betas are estimated against the equal-weight market and do not depend on the seed. Mean ±std over three seeds.
Estimate
Beta 20d
β^i
− 0.060 ±0.011
β^i×CVi
0.368 ±0.010
CVi
0.287
Trailing 4-day beta
0.434
Appendix
Table 14: Label-free beta estimates on the test period: daily RankIC with Beta 20d. β^i is the projection of the loading onto the date’s mean loading. Window statistics are computed on the stock’s own context window. The trailing four-day beta uses 5-minute returns of all stocks over the same window. Mean ±std over three seeds.
Learning effective representations helps characterize the structure and dynamics of equity markets from financial data with a low signal-to-noise ratio. Black-box deep models can capture complex patterns but may overfit sample noise and lack explicit economic structure. Meanwhile, classic linear financial models provide interpretable references, but their oversimplified assumptions leave non-linear signals uncaptured. To combine the strengths of these two directions, we propose Stock-JEPA, a joint-embedding predictive framework that learns predictable incremental revisions relative to a point-in-time financial prior. First, we leverage a low-complexity financial model to produce fixed statistics summarizing multi-horizon return and risk. A prior projector then maps these statistics into the target encoder's latent space as an anchor. Second, we design a context-conditioned revision predictor to estimate the future representation's predictable displacement from the anchor. Separate losses update the two branches: the anchor learns from prior statistics, while the revision captures additional predictable information from historical context. Third, we freeze all representation modules and train a downstream readout, evaluating its forecasts through cross-sectional ranking and portfolio performance. Theoretically, we prove that optimal revision reduces the prior anchor's expected squared error for the same future representation by exactly E[∥Δ∥22]. This non-negative gain is the expected squared magnitude of the additional signal predictable from historical context. Experimentally, Stock-JEPA outperforms 13 strong baselines across large-scale China and U.S. equity universes on 5 key evaluation metrics. Ablation studies and representation analysis further demonstrate the value of the learned revisions for representation learning in equity markets.
Yizhi Luo, Jiahe Yi, Jianhui Zhang +1
The Hong Kong University of Science and Technology · Shanghai Jiaotong University · Beijing Zhongguancun Academy
Building a world model requires a state representation useful for planning and decision-making---potentially over tasks unknown at training time. In the context of financial markets, planning and decision-making may require a model to reason about market-wide conditions, asset-specific expected returns, liquidity, volatility, and cross-asset relationships. Yet financial representation learning has largely been evaluated on individual predictive tasks, oftentimes on a single time period using comparatively narrow datasets. We address this through three primary contributions. First, we introduce Market-1T, a dataset containing nearly one trillion observations across U.S. equities from 2008 to 2025 at 1 Hz resolution. Second, we develop and implement a rigorous evaluation protocol. Third, we conduct a systematic large-scale study of financial representation learning, comparing 18 encoder-training strategies across nearly two decades of market regimes. We evaluate learned representations both by their predictive utility on common finance tasks and through probes of latent structure. We find that encoders with similar predictive performance can organize market state very differently. Collectively, we establish a foundation for training and evaluating financial market representations in support of world models such as DINO-WM, V-JEPA 2, and LeWM.
Humzah Merchant, Alec Guthrie, Simon Mahns +2
University of Chicago · Johns Hopkins University · Brown University
Financial time-series pretraining typically learns from masked observations, contrastive relations, or future outcomes---yet existing objectives struggle to simultaneously avoid future-supervision uncertainty and maintain return-prediction alignment. We propose DIVINE (DIVerse INdicator rEconstruction), a simple cross-market pretraining framework that reconstructs technical indicators from raw OHLCV history. Computed from observed price-volume history, technical indicators provide consistently defined supervision across markets while summarizing diverse market dynamics with established relevance to return prediction. Pretrained jointly on six-equity market datasets, DIVINE reconstructs 77 targets derived from 16 standard indicators and transfers only the learned encoder to downstream stock ranking. Across all six markets, DIVINE achieves the strongest average portfolio performance with a lightweight 0.05M-parameter encoder, outperforming pretraining baselines and matching or exceeding substantially larger financial foundation models, while remaining robust and data-efficient. Systematic analyses show that indicator diversity and market diversity provide complementary gains in transfer. Together, these results suggest that supervision design and cross-market diversity---rather than model scale---are the key drivers of strong, transferable financial representations.