Preprint
Article

This version is not peer-reviewed.

RNAD: A Regime-Normalized, Tail-Robust and Direction-Aware Loss for Deep Time-Series Forecasting

A peer-reviewed article of this preprint also exists.

Submitted:

03 November 2026

Posted:

04 November 2026

You are already at the latest version

Abstract
Financial time-series forecasting is commonly optimized with mean squared error (MSE), yet volatility clustering, heavy tails, and regime shifts make a fixed quadratic objective poorly suited. Existing robust losses improve resistance to extreme errors but lack adaptive volatility normalization and explicit directional supervision, while directional objectives such as Generalized Mean Absolute Directional Loss (GMADL) do not integrate volatility scaling, tail adaptation, and noise-aware directional weighting. We introduce Regime-Normalized Adaptive Directional Loss (RNAD), a model-agnostic training objective combining online volatility normalization via batch-scale estimates and an exponential moving average, tail-gated interpolation between quadratic and Charbonnier penalties, and noise-gated directional supervision that attenuates near-zero returns. RNAD requires no architectural changes, adds only a scalar state buffer, and incurs no inference-time overhead. Across nine financial assets, four architectures (PatchTST, iTransformer, N-HiTS, TimeXer), five losses, and three seeds under chronological splits, RNAD achieves the lowest mean RMSE for every backbone, reducing RMSE by 0.086–0.882 units versus the strongest non-RNAD baseline. It also improves MAE and directional accuracy (DA), with the lowest RMSE on all nine assets when paired with PatchTST. Four-step forecasts on US500 and USDJPY improve average DA by 2.39 and 1.41 percentage points, respectively, and reduce flattened RMSE on US500 by 0.3%. Ablations identify directional supervision as the main contributor, with tail adaptation and noise gating providing complementary, regime-dependent gains. Overall, adaptive, volatility-aware loss design offers a lightweight route to better numerical accuracy and directional consistency across financial forecasting architectures.
Keywords: 
;  ;  ;  ;  ;  

1. Introduction

Financial forecasting is difficult not because modern models lack capacity, but because the underlying process is unstable. Asset prices and returns exhibit time-varying volatility (Bollerslev, 1986; Engle, 1982), heavy-tailed shocks (Cont, 2001; Mandelbrot, 1963), and regime changes (Hamilton, 1989) that alter both the scale and the interpretation of forecast residuals. In short-horizon settings, especially at high frequency, a predicted return that gets the sign wrong can be less useful than a slightly less accurate predicted return with the correct sign. This makes the choice of training objective central: it must reflect not only the magnitude of the forecast residual, but also the market structure in which the forecast will be used.
Most forecasting systems, however, are still trained with mean-squared error or related fixed-scale regression losses. These objectives are convenient, but they implicitly assume a stable residual distribution and give the same treatment to all errors across regimes. Robust variants such as Huber or log-cosh reduce sensitivity to large residuals, yet they remain primarily magnitude-based and do not adapt to changing volatility (Huber, 1964; Saleh & Saleh, 2022). Directional criteria, on the other hand, directly capture sign correctness, but they are often used only as evaluation metrics or are applied uniformly even when the realized return is too small to carry clear economic meaning (Leitch & Tanner, 1991; Michańków et al., 2024; Pesaran & Timmermann, 1992). The resulting gap is narrow but important: existing objectives capture either robustness or direction, but not both in a way that responds to regime changes.
This paper addresses that gap with RNAD, a unified loss for regime-switching financial forecasting. The objective combines adaptive volatility normalization, a robust magnitude term, and a directional component that is weighted by the relevance of the observed return. The aim is to make the learning signal more consistent with the statistical structure of financial data while remaining compatible with standard forecasting backbones and requiring no architectural changes. Conceptually, RNAD treats volatility, outliers, and directional ambiguity as properties of the training objective rather than as separate preprocessing or post hoc evaluation issues.
We evaluate the proposed loss on nine minute-frequency assets spanning equities, foreign exchange, commodities, cryptocurrencies, and equity indices, using four forecasting backbones under strict chronological splits. Across this benchmark, RNAD achieves the lowest RMSE on every backbone while also improving MAE and directional accuracy. The ablation study further shows that the directional component contributes most of the improvement, while the tail-robust and noise-sensitive components provide smaller, regime-dependent gains. Taken together, these results suggest that loss design can materially influence both forecast accuracy and sign consistency in financial prediction.
The remainder of the paper reviews related work, develops the loss function, and presents the experimental analysis.

3. The RNAD Loss Function

This section formalizes the loss used throughout the paper. After defining the forecasting notation, it develops the volatility-normalized magnitude term, the gated directional term, and the final batch objective. The presentation is organized to make the optimization geometry and the runtime statefulness of RNAD explicit before turning to implementation details.

3.1. Notation and Problem Formulation

Consider one-step-ahead forecasting of a univariate close-price series { y t } t = 1 T , where y t ∈ R > 0 denotes the close price at time t. For an example whose input ends at time t b , the forecasting model produces
y ^ b = f θ ( x t b − L + 1 : t b ) , y b = y t b + 1 , y prev , b = y t b .
The input therefore ends at t b , while the target is the next close and is not part of the input. In implementation, both the input features and the target close series are standardized using training-split z-score statistics before the loss is evaluated; predictions are inverse-transformed only for raw-scale reporting.
For each mini-batch example b, define:
e b = y ^ b − y b , ( forecast residual ) ,
r b = y b − y prev , b , ( realized one - step return ) ,
r ^ b = y ^ b − y prev , b , ( predicted one - step return ) .
Here b ∈ { 1 , … , B } indexes examples within a mini-batch of size B. All three quantities in Eqs. (1)–(3) are computable from the current prediction and the target window; the model input itself contains information only through t b , so the loss introduces no look-ahead into forecasting.

3.2. Adaptive Volatility Normalization

A shared volatility scale s > 0 underlies all subsequent computations. It blends two complementary estimators: a batch-wise instantaneous estimate that responds quickly to volatility spikes, and an EMA that retains regime-level memory across gradient steps.

Batch-wise scale.

Given a mini-batch of realized returns { r b } b = 1 B , the batch-wise scale is defined as
s batch = 1 B ∑ b = 1 B r b 2 + ε , ε = 10 − 8 .
This is the root-mean-square (RMS) of realized returns, which approximates the conditional standard deviation under zero-mean innovations. The regularizer ε prevents division by zero during sustained flat-price periods.

EMA scale.

To capture regime-level volatility persistence, a running exponential moving average is maintained:
s ˜ = ρ s ¯ + ( 1 − ρ ) s batch , ρ = 0.985 ,
where s ¯ is the buffer value at the start of the forward pass and is initialized to 1.0 . The effective memory window spans approximately 1 / ( 1 − ρ ) = 67 gradient steps, which is sufficient to track macro-level regime transitions while remaining responsive to sustained volatility shifts.

Composite scale.

The final scale is the geometric mean of the two estimates:
s = max s batch s ˜ + ε , ε .
The outer maximum is the implementation’s divisor guard. Taking the geometric mean ensures that s responds promptly to volatility spikes via s batch while being anchored to the prevailing regime level via s ˜ , without tracking transient noise. The normalized quantities used throughout are:
u b = e b s , z b = r b s , z ^ b = r ^ b s ,
where u b is the volatility-normalized residual and z b , z ^ b are the normalized realized and predicted returns, respectively. This normalization removes the spurious coupling between gradient magnitude and asset price level; without it, high-priced assets can generate gradients much larger than low-priced ones, making a single hyperparameter configuration less transferable across asset classes (cf. Section 3.6).
Implementation note. The codebase standardizes all input features and the target close series using training-split statistics before evaluating the loss; the notation above is written on that standardized target scale, and raw prices are restored only for reporting.
Remark 1. 
The EMA update in Eq. (5) is applied only during training. During validation and testing, s ˜ is set to the frozen buffer s ¯ ; the current batch scale is still used for normalization, but it does not update persistent state. Checkpoint- resume pipelines should therefore persist s ¯ alongside model parameters.

3.3. Tail-Robust Magnitude Term

The magnitude component smoothly interpolates between a quadratic kernel and the sub-linear Charbonnier penalty used in robust estimation (Charbonnier et al., 1997; Huber, 1964). The former preserves sensitivity for well-predicted samples, whereas the latter limits the growth of the penalty during tail events.

Tail gate.

Let σ ( x ) = ( 1 + e − x ) − 1 denote the logistic sigmoid. The interpolation weight is
g b = σ | u b | − τ mag τ temp + ε ,
where τ mag = 1.1 is the transition midpoint in units of s and τ temp = 0.3 controls the sharpness of the transition. Tail gating prevents rare shock events from dominating optimization: unlike a hard threshold, the smooth transition preserves gradient informativeness in the interior while bounding gradient norms in the tails (cf. Section 3.6). When | u b | ≪ τ mag , we have g b ≈ 0 (quadratic regime); when | u b | ≫ τ mag , g b ≈ 1 (Charbonnier regime). The logistic transition avoids the gradient discontinuity of the Huber threshold.

Kernel definitions.

The quadratic and Charbonnier kernels are:
l b L 2 = 1 2 u b 2 ,
l b charb = u b 2 + ε c 2 − ε c , ε c = 10 − 3 .
The offset ε c ensures differentiability at u b = 0 . As | u b | → ∞ , l b charb ≈ | u b | (linear growth), whereas l b L 2 = 1 2 u b 2 grows without bound, amplifying the influence of extreme residuals.

Blended magnitude loss.

The final magnitude term is:
l b mag = ( 1 − g b ) l b L 2 + g b l b charb .
Because the gates depend on | u b | and | z b | , the full objective is piecewise smooth rather than globally C ∞ ; it is differentiable almost everywhere, with non-differentiable points only at u b = 0 and z b = 0 . Away from those measure-zero points, the derivative with respect to u b transitions from approximately u b near the origin to approximately sgn ( u b ) in the tails, bounding the influence of tail observations on parameter updates.

3.4. Directional Term with Noise Gating

The directional component penalizes sign disagreement between the predicted and realized return, following the motivation for direction-sensitive forecast evaluation (Leitch & Tanner, 1991; Pesaran & Timmermann, 1992). A smooth gate downweights the penalty on economically small returns.

Smooth sign encoding.

Both returns are mapped to ( − 1 , + 1 ) via a scaled hyperbolic tangent:
d b true = tanh ( β z b ) , d b pred = tanh ( β z ^ b ) , β = 4.5 .
The sharpness parameter β controls the transition width around zero. At β = 4.5 , a normalized return | z b | = 0.5 is encoded as | tanh ( 2.25 ) | ≈ 0.978 , rendering the encoding near-binary for moderate returns while remaining differentiable at the origin.

Directional alignment score.

The per-sample alignment is:
a b = d b true · d b pred ∈ [ − 1 , 1 ] .
Perfect directional agreement ( z ^ b and z b share sign) gives a b → + 1 ; perfect disagreement gives a b → − 1 .

Noise gate.

A smooth sigmoid gate downweights directional supervision when the realized return is negligible relative to the current volatility:
w b = σ | z b | − τ noise τ noise , temp + ε , τ noise = 0.1 .
This gate focuses directional supervision on economically meaningful returns; near-zero returns can otherwise contribute noisy, potentially misleading sign gradients that destabilize learning in low-volatility regimes (cf. Section 3.6). The gate temperature is fixed at τ noise , temp = 0.04 . Both constants are expressed in normalized units and are held fixed across all assets; no per-asset calibration is applied. Smaller returns receive continuously smaller directional weights rather than a binary on/off suppression.

Directional penalty.

Using softplus for numerical stability, the directional loss is:
l b dir = w b · softplus − κ a b , softplus ( x ) = ln ( 1 + e x ) , κ = 4.5 .
For perfect alignment ( a b = 1 ): l b dir = softplus ( − 4.5 ) ≈ 0.011 . For perfect disagreement ( a b = − 1 ): l b dir = softplus ( 4.5 ) ≈ 4.51 . This asymmetry strongly penalizes directional errors while imposing negligible cost on correct predictions, creating a well-defined decision boundary around the sign of z ^ b .

3.5. Full Objective

The per-sample RNAD loss is:
L b R N A D = l b mag + λ dir l b dir , λ dir = 0.5 ,
and the mini-batch training objective averages over the batch:
L R N A D = 1 B ∑ b = 1 B L b R N A D .
The EMA buffer s ¯ is the only persistent state. During training, it is updated once per forward pass before the loss is returned, and the current batch is normalized using the post-update estimate s ˜ . Consequently, the volatility estimate used in Eqs. (4)–(16) incorporates the current batch statistics and the updated EMA state. During evaluation, the EMA state remains unchanged. Figure 1 summarizes this loss architecture: it shows how the shared normalization feeds the magnitude and directional branches before their combination, but it is not intended to specify execution order.
During validation and testing the buffer is frozen, so no evaluation- set statistics feed back into the training state (cf. Remark 1). All remaining quantities in Eqs. (8)–(16) are computed solely from the current batch. Algorithm 1 supplies the corresponding computational execution order, including coercion, normalization, branch evaluation, loss assembly, and the training/validation state transition. A stateless diagnostic variant of the loss is documented in Section 5. Table 5 lists all loss-specific constants, and Section 5 summarizes the full backbone and training settings.
Algorithm 1. RNAD forward pass (one mini-batch) Implements Eqs. (4)–(17). Phases 0–4 are stateless; only Phase 5 mutates persistent state, and only in training mode.
Input:
predictions y ^ ∈ R B ; targets y ∈ R B ; previous targets y prev ∈ R B
Input:
constants Θ = λ dir , τ mag , τ temp , ε c , β , κ , τ noise , τ noise , temp , ρ , ε (Table 5)
Input:
EMA volatility buffer s ¯ ∈ R > 0 , initialised to 1 and serialised in checkpoints
Output:
scalar loss L RNAD ; (training mode only) refreshed state s ¯
⁠
     Phase 0 — coercion and state initialisation
  1:
      y ← cast ( y , dtype ( y ^ ) , device ( y ^ ) ) ; y prev ← cast ( y prev , dtype ( y ^ ) , device ( y ^ ) )
  2:
if s ¯ is undefined then
  3:
     s ¯ ← 1
  4:
end if
⁠
     Phase 1 — residuals and return-like deltas
  5:
e b ← y ^ b − y b                                                             ▹Eq. (1)
  6:
r b ← y b − y prev , b                                                         ▹Eq. (2)
  7:
r ^ b ← y ^ b − y prev , b                                                        ▹Eq. (3)
⁠
      all three are computed on the standardized target scale; predictions are inverse-transformed only for reporting
⁠
     Phase 2 — composite volatility scale
  8:
s batch ← 1 B ∑ b = 1 B r b 2 + ε                                                      ▹Eq. (4)
  9:
if training mode then
10:
     s ˜ ← ρ s ¯ + ( 1 − ρ ) s batch                                                      ▹Eq. (5)
11:
else
12:
     s ˜ ← s ¯                                                                  ▹frozen buffer; no evaluation-set statistics
13:
end if
14:
s ← max s batch · s ˜ + ε , ε                                                      ▹Eq. (6)
⁠
      the inner max is a divisor guard; s ˜ is treated as a constant, so no gradient flows through s
⁠
     Phase 3 — volatility-normalised quantities
15:
u b ← e b / s ; z b ← r b / s ; z ^ b ← r ^ b / s                                             ▹Eq. (7)
⁠
     Phase 4a — tail-gated magnitude term
16:
for b = 1 , … , B do
17:
     g b ← σ | u b | − τ mag τ temp + ε                                                        ▹Eq. (8)
18:
     l b L 2 ← 1 2 u b 2                                                                   ▹Eq. (9)
19:
     l b charb ← u b 2 + ε c 2 − ε c                                                            ▹Eq. (10)
20:
     l b mag ← ( 1 − g b ) l b L 2 + g b l b charb                                                       ▹Eq. (11)
21:
end for
22:
L mag ← 1 B ∑ b = 1 B l b mag
⁠
     Phase 4b — noise-gated directional term
23:
for b = 1 , … , B do
24:
     w b ← σ | z b | − τ noise τ noise , temp + ε                                                     ▹Eq. (14)
25:
     d b true ← tanh ( β z b ) ; d b pred ← tanh ( β z ^ b )                                            ▹Eq. (12)
26:
     a b ← d b true · d b pred ∈ [ − 1 , + 1 ]                                                    ▹Eq. (13)
27:
     l b dir ← w b · softplus ( − κ a b )                                                  ▹Eq. (15)
28:
end for
29:
L dir ← 1 B ∑ b = 1 B l b dir
⁠
     Phase 5 — assembly and state commit
30:
L RNAD ← L mag + λ dir L dir                                                   ▹Eq. (17)
31:
if training mode then
32:
     s ¯ ← detach ( s ˜ )                                                            ▹state write; detached from the autograd graph
33:
end if
34:
return L RNAD , s ¯
Algorithm 1 performs O ( B ) elementwise operations per call, holds a single scalar buffer, and adds no synchronisation barrier; it therefore introduces no measurable inference overhead and negligible training overhead (Table 12).
With that execution order established, Figure 2 visualizes the optimization geometry produced by the combined loss: the interaction between normalized error and directional alignment.

3.6. Mechanistic Rationale

Figure 2 provides the geometric view of the objective, while the component analysis below explains which failure mode of standard losses each part of RNAD targets.
  • Volatility normalization (Eq. 6, Phase 2 of Algorithm 1) removes the spurious coupling between gradient magnitude and asset price level. Without normalization, high-priced assets (BTCUSD ∼$21,000) generate gradients orders of magnitude larger than low-priced assets (AUDUSD ∼0.65), preventing a single hyperparameter configuration from being applicable across asset classes.
  • Tail gating (Eq. 8, Phase 4a) prevents rare shock events from dominating optimization. Unlike Huber’s hard threshold, the smooth transition preserves gradient informativeness within the interior of the residual distribution while bounding gradient norms in the tails, improving robustness without sacrificing sensitivity.
  • Directional learning (Eq. 15, Phase 4b) explicitly incentivizes sign correctness, addressing MSE’s structural inability to enforce directional accuracy (Leitch & Tanner, 1991; Pesaran & Timmermann, 1992). As analyzed in Section 8.2, this coupling between sign and magnitude objectives also reduces RMSE.
  • Noise gating (Eq. 14, Phase 4b) focuses directional supervision on economically meaningful returns. Near-zero returns contribute noisy, potentially misleading sign gradients that can destabilize learning in low-volatility regimes such as foreign exchange markets (see ablation results in Table 10).
The experimental setup required to evaluate these mechanisms is described in the following section.

4. Experimental Setup

This section specifies the data universe, preprocessing pipeline, baselines, backbone configurations, and metrics used to evaluate RNAD. These choices are central because the loss is designed to operate across assets with very different volatility scales under strict chronological validation. The setup is therefore described in enough detail to make the benchmark reproducible and to clarify how the reported gains should be interpreted.

4.1. Datasets and Assets

We evaluate on nine financial assets spanning five market classes (Table 1): equities (AAPL, DAX30, FTSE100), foreign exchange (AUDUSD, USDJPY), cryptocurrencies (BTCUSD, ETHUSD), equity indices (US500), and commodities (XAUUSD). All series consist of minute-frequency OHLCV data. Total sample sizes range from 15,673 (AAPL) to 68,536 (USDJPY), spanning multiple market regimes and volatility cycles. Realized volatility proxies (training-set RMS returns) vary by two orders of magnitude across assets, from USDJPY ( 9.6 × 10 − 4 ) to ETHUSD ( 8.5 × 10 − 3 ), providing strong motivation for the cross-asset normalization mechanism.

Chronological split protocol.

All splits are strictly chronological with no shuffling: 80 % training, 10 % validation, 10 % test. This protocol eliminates look-ahead bias, preserves temporal autocorrelation structure, and evaluates generalization to the most recent—and typically the most distributional-shift-prone—period of each series, following established best practice (Makridakis et al., 2022; Zeng et al., 2023).

4.2. Feature Engineering and Preprocessing

All models receive the same causally valid feature set, computed exclusively from observations available at or before time t: OHLCV bars; MACD (line, signal, histogram); RSI-14; EMA-9; and Bollinger Bands (SMA-20, upper band, lower band). The 13-column feature vector is ordered as [open, high, low, close, volume, macd, macd_signal, macd_hist, rsi_14, ema_9, bb_middle, bb_upper, bb_lower]. The input window consists of the L = 24 most recent observations x t − 23 , … , x t , and the model is trained to forecast the next close price y t + 1 (i.e., the close at time t + 1 , which is not part of the input). Thus the target is the future close price, while the close channel at index 3 in the input represents the already observed close prices up to time t.
Both the input features and the target close-price series are standardized using z-score statistics ( μ , σ ) computed on the training split alone and applied identically to the validation and test splits. The fitted target scaler is retained to inverse-transform predictions before computing raw-scale RMSE, MAE, and directional accuracy. The forecasting window is fixed at L = 24 input steps with an output horizon of H = 1 step ahead across all experiments.

4.3. Baseline Loss Functions

We compare RNAD against four established loss functions. All baselines are evaluated under the identical chronological split protocol, training budget, optimisation settings, and evaluation metrics as RNAD. To guarantee a fair comparison, every baseline hyperparameter that can materially influence performance was optimised on the validation split using the same procedure.
  • MSE: Standard mean-squared error; contains no tunable hyperparameters.
  • Huber: Fixed-threshold interpolation between quadratic and linear regimes (Huber, 1964). The threshold δ was tuned via grid search over { 0.1 , 0.5 , 1.0 , 2.0 , 5.0 } ; the value δ = 1.0 yielded the lowest validation RMSE across the majority of backbone–asset combinations and was therefore adopted uniformly.
  • log-cosh: Smooth approximation to MAE with bounded second derivatives (Saleh & Saleh, 2022). Contains no tunable hyperparameters.
  • GMADL: Combined magnitude and directional loss without volatility normalization or noise gating (Michańków et al., 2024). The weighting parameters were optimised for the current datasets via a grid search over α ∈ { 0.1 , 0.5 , 1.0 } and β ∈ { 1.0 , 5.0 , 10.0 } using the validation split, selecting the configuration with lowest validation RMSE. The optimal values were α = 0.5 , β = 5.0 .

4.4. Forecasting Backbones

The four evaluation backbones are:
  • PatchTST (Nie et al., 2023): Patch-based transformer with channel-independent self-attention and RevIN normalization; L = 24 , patch length 16, stride 8 (two patches), d model = 128 , d ff = 256 , 4 heads, 3 encoder layers, dropout 0.05 , fc-dropout 0.05 , and head-dropout 0.0 .
  • iTransformer (Liu et al., 2024): Inverted transformer with a 24-step linear inverted embedding, d model = 128 , d ff = 256 , 8 heads, 3 encoder layers, dropout 0.05 , and GELU activation.
  • N-HiTS (Challu et al., 2023): Neural hierarchical interpolation with 3 stacks, 2 blocks per stack, 2 MLP layers per block, n hidden = 256 , max pooling, kernel sizes [ 2 , 2 , 2 ] , downsampling factors [ 4 , 2 , 1 ] , linear interpolation, dropout 0.05 , ReLU activation, no batch normalization, and no shared weights.
  • TimeXer (Wang et al., 2024): Patch-based transformer with patch length 8 (three non-overlapping target patches over L = 24 ), d model = 128 , d ff = 256 , 8 heads, 2 encoder layers, dropout 0.05 , factor 5, GELU activation, and use_norm=true.
All backbones forecast the close channel (index 3) from the same 13-feature input vector. These architectures span transformer, inverted-attention, hierarchical interpolation, and patch-based cross-attention paradigms, providing a rigorous test of loss-level generalization across diverse inductive biases.

4.5. Multi-Step Forecasting Setup

To evaluate whether RNAD’s benefits persist in multi-step prediction, PatchTST was trained to forecast the next H = 4 close-price observations on the US500 and USDJPY assets. The input window remains L = 24 time steps, and the model outputs a vector of four predictions. The output horizon is appended to the backbone’s projection layer without any further architectural changes. Training uses the same optimization settings, batch size ( B = 64 ), seed (178), and 80/10/10 chronological split described in Section 4 and Section 5. The validation-set RMSE is computed over the flattened set of all 4 × N validation samples, and the checkpoint with the best flattened validation RMSE is retained for testing.
Performance is reported as per-step root mean squared error (RMSE) and directional accuracy (DA), together with the flattened RMSE computed over all four steps and the arithmetic mean of the four horizon- specific directional accuracies.

5. Implementation Details and Reproducibility

To ensure reproducibility, we fix the exact backbone, optimization, and deterministic settings used throughout the benchmark. The feature order and target index are identical across all runs, and the same configuration file drives every experiment.

5.1. Backbone Configurations

To ensure a fair comparison across architectures, all forecasting backbones were trained under a unified experimental setting. Each backbone received the same 24-step historical input window constructed from 13 standardized market features and was tasked with predicting the closing price channel (feature index 3). The complete backbone-specific hyperparameter settings are summarized in Table 2.
Unless otherwise specified, all remaining hyperparameters follow the default settings provided in the original model implementations. Additional constructor-level options were retained to preserve architectural consistency with the reference implementations. For instance, PatchTST additionally employs attn_dropout=0, padding_patch=None, affine=true, subtract_last=false, pre_norm=false, head_type=flatten, and max_seq_len=1024.

5.2. Hardware and Training Duration

All experiments were run on a single NVIDIA T4 Tensor Core GPU (16 GB). The full benchmark—4 backbones × 5 losses × 3 seeds × 9 assets (including validation and final test evaluation)—required approximately 105 hours of total wall-clock time. The random seeds are { 178 , 956 , 1045 } for the main benchmark and 178 for all ablation studies, consistent with Table 4. Complete hardware, software, and reproducibility details are summarized in Table 3.
Table 3. Experimental environment and reproducibility summary.
Table 3. Experimental environment and reproducibility summary.
Parameter Value / Description
Hardware
GPU NVIDIA T4 Tensor Core (16 GB)
CPU 4 vCPUs (Intel Xeon)
System RAM 16 GB
Software Environment
Operating system Ubuntu 22.04 LTS
Python 3.11
PyTorch 2.7.1
CUDA Toolkit 12.8
cuDNN 9.8
Training Efficiency
Total wall-clock time ∼105 h (4 backbones × 9 assets × 5 losses × 3 seeds = 540 runs)
Reproducibility
Data split Chronological 80/10/10; no shuffling of time order
Determinism Fixed Python, NumPy, PyTorch, and CUDA seeds; cuDNN deterministic
Table 4. Shared optimization and reproducibility settings for the benchmark. The table records the training budget, batching, checkpointing, seed control, and determinism settings used across all runs.
Table 4. Shared optimization and reproducibility settings for the benchmark. The table records the training budget, batching, checkpointing, seed control, and determinism settings used across all runs.
Setting Value Notes
Optimizer Adam PyTorch defaults: β 1 = 0.9 , β 2 = 0.999 , ϵ = 10 − 8 , weight decay = 0 , AMSGrad disabled
Learning rate 5 × 10 − 4 Constant across all runs; no scheduler, warmup, or gradient clipping
Epochs 50 Full training budget; best validation-RMSE checkpoint restored after training
Batch size 64 All loaders preserve the original temporal order; no shuffling is applied.
Checkpointing every epoch checkpoint_every=1; resume=true
Seeds { 178 , 956 , 1045 } Main benchmark uses three seeds; ablation studies use seed 178
Determinism true Python, NumPy, PyTorch, and CUDA seeds are fixed; cuDNN deterministic paths are enabled

5.3. Optimization and Deterministic Training

All models are trained with Adam using the PyTorch defaults ( β 1 = 0.9 , β 2 = 0.999 , ϵ = 10 − 8 , zero weight decay, no AMSGrad) and a fixed learning rate of 5 × 10 − 4 . No learning-rate schedule, warmup, or gradient clipping is applied. Training runs for 50 epochs with batch size B = 64 and num_workers = 0.Checkpoints are saved every epoch, resume=true is enabled, and the checkpoint with the best validation RMSE is restored after training.

5.4. RNAD Loss Hyperparameters

RNAD uses fixed scalar controls across all runs; the loss contains no per-asset calibration or learned parameters. The EMA volatility buffer is initialized to 1.0, updated online, and serialized in checkpoints. Table 5 lists the loss-specific constants.
Table 5. Fixed RNAD hyperparameters used in all experiments. These constants control the magnitude gate, directional gate, and EMA volatility state, and are held constant across assets and backbones.
Table 5. Fixed RNAD hyperparameters used in all experiments. These constants control the magnitude gate, directional gate, and EMA volatility state, and are held constant across assets and backbones.
Parameter Value Role
λ dir 0.5 Directional penalty weight (Eq. 16)
τ mag 1.1 Tail-gate transition midpoint (Eq. 8)
τ temp 0.3 Tail-gate transition sharpness (Eq. 8)
ε c 10 − 3 Charbonnier smoothness offset (Eq. 10)
β 4.5 Tanh sharpness for sign encoding (Eq. 12)
κ 4.5 Softplus margin slope (Eq. 15)
τ noise 0.10 Smooth noise-gate center (Eq. 14)
τ noise , temp 0.04 Noise-gate temperature (Eq. 14)
ρ 0.985 EMA momentum, ≈ 67 -step window (Eq. 5)
ε 10 − 8 Numerical stability constant

5.5. Hyperparameter Selection

The fixed hyperparameters reported in Table 5 were selected based on the mathematical behaviour of the proposed RNAD objective, the statistical characteristics of financial time-series, and preliminary empirical tuning during method development. Rather than performing extensive hyperparameter optimization for each dataset, we adopted a single set of values across all experiments to ensure a fair and consistent evaluation of the proposed loss.

Tail-gate transition midpoint τ mag = 1.1 .

The parameter τ mag determines when RNAD gradually transitions from the quadratic magnitude penalty to the more robust Charbonnier penalty. Setting τ mag = 1.1 allows normal prediction errors to be optimized using the quadratic loss while reducing the influence of unusually large residuals caused by extreme market movements. Smaller values would suppress useful gradients for ordinary samples, whereas substantially larger values would reduce the robustness of the loss to outliers.

Sign-encoding sharpness and margin slope β = κ = 4.5 .

The parameters β and κ control the smoothness of the differentiable directional supervision. The value 4.5 provides a sufficiently sharp approximation to the binary sign function while maintaining stable gradients for gradient-based optimization. Using the same value for both parameters also keeps the sign encoding and directional margin consistent within the loss formulation.

EMA momentum ρ = 0.985 .

The exponential moving average coefficient ρ controls how quickly the volatility normalization adapts to changing market conditions. A value of 0.985 provides a balance between stability and responsiveness by smoothing short-term fluctuations while remaining capable of tracking gradual changes in market volatility. Larger values would adapt more slowly, whereas smaller values would make the normalization more sensitive to batch-level noise.

Directional weight λ dir = 0.5 .

The weighting coefficient λ dir balances the magnitude regression objective and the directional supervision objective. Setting λ dir = 0.5 encourages the model to improve directional predictions without allowing the directional component to dominate the overall optimization. This value provided a balanced trade-off between regression accuracy and directional performance during preliminary development.

Remaining hyperparameters.

The remaining parameters, τ temp and τ noise , govern the smoothness of the gating mechanisms: τ temp controls the sharpness of the tail-gate transition (Eq. 8), while τ noise sets the centre of the noise gate (Eq. 14). Their selected values provide smooth transitions between regimes without introducing abrupt changes in the optimisation landscape. The same hyperparameter configuration was maintained across all datasets and forecasting models to ensure consistency and avoid dataset-specific tuning.
Overall, the RNAD hyperparameters were selected to provide a stable compromise between robustness to heavy-tailed financial data, accurate magnitude prediction, and reliable directional supervision. Importantly, the same configuration was used throughout all experiments reported in this paper.

5.6. Evaluation Metrics

Predictions are inverse-transformed to raw close prices using the fitted target scaler before RMSE, MAE, and directional accuracy (DA) are computed.

Primary — RMSE.

Root Mean Squared Error on the raw price scale (lower is better):
RMSE = 1 | T | ∑ t ∈ T y ^ t − y t 2 .

Primary — Directional Accuracy (DA).

The proportion of non-zero returns for which the predicted return sign matches the realized sign (higher is better):
DA = 1 | T | ∑ t ∈ T 1 sgn ( r ^ t ) = sgn ( r t ) , T = t : r t ≠ 0 .

Secondary.

Mean Absolute Error (MAE) and ordinal RMSE rank across losses within each backbone are reported as secondary diagnostics to assess consistency.
All results are expressed as mean ± standard deviation over the three benchmark seeds.

6. Results

This section presents the benchmark outcomes in three layers: aggregate backbone-level comparisons, backbone-wise averages, and per-asset behavior. The ordering is deliberate, moving from the broadest evidence to the more granular market-specific patterns that explain it. This structure helps distinguish loss-level effects that are consistent across architectures from those that are asset-dependent.

6.1. Main Benchmark: Backbone–Loss Comparison

Table 6 compares the five loss functions across all four backbones, averaged over nine assets and three random seeds. RNAD achieves the lowest mean RMSE on every backbone while also improving MAE and directional accuracy (DA). Because RMSE is scale-dependent, these cross-asset averages are influenced more heavily by higher-priced assets; the per-asset analysis in Table 8 therefore provides essential context.
Compared with the strongest non-RNAD baseline within each backbone, RNAD yields consistent RMSE improvements. The largest reductions are observed on PatchTST ( Δ RMSE = − 0.882 ) and TimeXer ( − 0.658 ), followed by iTransformer ( − 0.681 ). On N-HiTS, the gain is smaller but remains positive ( Δ RMSE = − 0.086 ). In addition, RNAD achieves the highest DA across all four backbones, indicating improved directional forecasting performance alongside lower forecast residual.

6.2. Cross-Architecture Consistency

Table 7 aggregates metrics by backbone across all nine assets and three seeds. RNAD achieves the best average RMSE on all four architectures and reduces RMSE relative to MSE by 0.09–0.89 units. Standard deviations of per-seed RMSE on transformer-family backbones remain small ( σ ≤ 0.07 ; Table 6), indicating that the gains are stable across random seeds.
The contrast between transformer-family backbones and N-HiTS suggests that backbone-level regime awareness already addresses part of the same failure mode. RNAD is therefore most useful when paired with an expressive backbone that lacks built-in regime awareness.

6.3. Per-Asset Analysis

Table 8 reports per-asset RMSE and DA on PatchTST. RNAD achieves the lowest RMSE on all nine assets, indicating that the gains are not concentrated within a single market class. The largest proportional reductions occur on BTCUSD ( − 6.69 units, − 1.56 % ) and AAPL ( − 0.133 units, − 7.6 % ), the two most volatile assets in Table 1. Directional accuracy gains are smaller and more heterogeneous, ranging from + 0.17 percentage points (USDJPY) to + 2.29 pp (BTCUSD).
Table 8. PatchTST results by asset, averaged over three seeds ( { 178 , 956 , 1045 } ). RMSE and DA (%) are reported for each loss; bold and underlining mark the best and second-best values within each metric for each asset.
Table 8. PatchTST results by asset, averaged over three seeds ( { 178 , 956 , 1045 } ). RMSE and DA (%) are reported for each loss; bold and underlining mark the best and second-best values within each metric for each asset.
MSE Huber log-cosh GMADL RNAD
Asset RMSE DA RMSE DA RMSE DA RMSE DA RMSE DA
AAPL 1.75 48.62 1.73 48.43 1.71 49.25 1.67 50.14 1.61 50.75
AUDUSD 7.97e-4 49.61 7.97e-4 49.68 7.97e-4 49.74 8.02e-4 49.67 7.90e-4 50.56
BTCUSD 427.90 50.01 427.92 49.98 427.90 50.00 428.12 51.25 421.21 52.30
DAX30 54.93 50.62 54.93 50.62 54.93 50.59 54.90 50.17 54.50 51.58
ETHUSD 24.04 50.29 24.03 50.21 24.03 50.28 24.25 50.48 23.75 51.08
FTSE100 15.26 50.68 15.26 50.66 15.27 50.51 15.34 50.86 15.06 51.17
US500 14.47 49.98 14.47 49.85 14.47 49.93 14.49 50.29 14.29 50.26
USDJPY 0.18 50.73 0.18 50.51 0.18 50.76 0.18 50.28 0.18 50.89
XAUUSD 8.47 49.93 8.47 50.23 8.47 49.73 8.47 49.86 8.41 50.11

6.4. Multi-Step Forecasting Results

To evaluate whether RNAD’s benefits extend beyond one-step-ahead prediction, we trained PatchTST to forecast H = 4 steps on two representative assets, US500 and USDJPY, under the configuration described in Section 4.5. Training was run for 30 epochs (instead of the 50 used in the main benchmark) to keep the experiment lightweight; the checkpoint with the lowest flattened validation RMSE was retained. Table 9 reports the per-step RMSE and directional accuracy (DA), the flattened RMSE over all predictions, and the average DA across the four horizons.
Key observations:
  • RNAD substantially improves directional accuracy. Across both assets and all four horizons, RNAD achieves a higher per-step DA than MSE, with an average advantage of 2.39 percentage points on US500 and 1.41 pp on USDJPY. The gain is particularly striking at intermediate horizons ( H = 2 , 3 ) on US500, where DA improves by more than 2 pp.
  • RNAD yields a meaningful RMSE reduction on US500. The flattened RMSE drops from 22.447 (MSE) to 22.377 (RNAD), an improvement of 0.070 units ( − 0.3 % ). This demonstrates that the directional penalty not only sharpens sign predictions but also tangibly reduces overall scale error, consistent with the directional–magnitude coupling discussed in Section 8.2.
  • On USDJPY, RNAD slightly surpasses MSE in RMSE. The flat RMSE is marginally lower for RNAD ( 0.288 vs. 0.290 ), confirming that the loss remains competitive in a low-volatility FX regime without sacrificing scale accuracy.
  • Performance decays gracefully with horizon. As expected, RMSE increases and DA generally declines with lead time for both losses. However, RNAD’s relative advantage persists across all steps, indicating that the learned directional signal remains useful even four steps ahead.
These multi-step results confirm that RNAD generalises beyond the one-step setting without any architectural modification. The directional component remains the primary driver of improvement, yielding better sign consistency and, on the equity index, a measurable reduction in RMSE.

7. Ablation Analysis

To assess the individual contribution of each component, we train PatchTST with seed 178 under four conditions, removing one mechanism at a time from the full RNAD objective:
  • FULL: complete RNAD loss (Eq. 16);
  • NO DIR: directional term removed ( λ dir = 0 );
  • NO TAIL: gated magnitude replaced by a pure quadratic ( g b ≡ 0 in Eq. 11);
  • NO NOISE: noise gate disabled ( w b ≡ 1 in Eq. 14; directional term always active).
Table 10 reports results on four representative assets: a high-volatility cryptocurrency (BTCUSD), a low-volatility foreign exchange pair (AUDUSD), an equity index (US500), and a commodity (XAUUSD).
Table 10. Controlled ablation on PatchTST (seed 178). Δ RMSE and Δ DA are computed relative to FULL for each asset. Positive Δ RMSE indicates degradation when a component is removed. AUDUSD values are reported in scientific notation.
Table 10. Controlled ablation on PatchTST (seed 178). Δ RMSE and Δ DA are computed relative to FULL for each asset. Positive Δ RMSE indicates degradation when a component is removed. AUDUSD values are reported in scientific notation.
Asset Variant RMSE Δ RMSE MAE DA Δ DA
AUDUSD FULL 7.9020 × 10 − 4 ̲ — 5.3505 × 10 − 4 ̲ 50.84 —
NO DIR 7.9655 × 10−4 +6.34 × 10−6 5.4031 × 10−4 50.24 − 0.60
NO TAIL 7.9142 × 10−4 +1.22 × 10−6 5.3625 × 10−4 50.86 ̲ + 0.01
NO NOISE 7.8971 × 10 − 4 −4.9 × 10−7 5.3444 × 10 − 4 51 . 15 + 0.31
BTCUSD FULL 422.61 — 284.04 ̲ 51.67 —
NO DIR 425.32 + 2.71 286.09 51 . 74 + 0.06
NO TAIL 421 . 87 − 0.74 283 . 54 51.51 − 0.16
NO NOISE 422.42 ̲ − 0.18 284.27 51.72 ̲ + 0.05
US500 FULL 14 . 33 — 8.01 ̲ 49.57 —
NO DIR 14.33 ̲ + 0.01 8.06 ̲ 48.60 − 0.98
NO TAIL 14.35 + 0.02 8.03 50 . 27 + 0.70
NO NOISE 14.34 + 0.01 8 . 00 49.97 ̲ + 0.40
XAUUSD FULL 8.41 ̲ — 5.56 50.54 ̲ —
NO DIR 8.48 + 0.07 5.60 50.43 − 0.11
NO TAIL 8.42 + 0.01 5 . 56 50 . 90 + 0.35
NO NOISE 8 . 41 − 0.00 5.56 ̲ 50.39 − 0.15

Directional learning (NO DIR). 

Removing the directional penalty consistently degrades RMSE (mean Δ RMSE = + 0.70 across the four assets) and reduces DA on three of the four assets. This indicates that directional supervision contributes to both magnitude accuracy and directional accuracy, as discussed in Section 8.2. The effect is largest on BTCUSD ( Δ = + 2.71 ), where economically significant directional returns are most frequent.

Tail gating (NO TAIL). 

The contribution of tail gating is asset-dependent. On BTCUSD, removing the tail gate yields a marginal RMSE improvement ( − 0.74 ), suggesting that τ mag = 1.1 transitions slightly too aggressively for BTCUSD’s heavier tails. On AUDUSD, US500, and XAUUSD, the full loss is preferred over NO TAIL. These results indicate that τ mag is the parameter most sensitive to asset-specific tail behavior.

Noise gating (NO NOISE). 

The effects of noise gating are mixed and generally modest. On AUDUSD and BTCUSD, disabling the gate slightly improves RMSE ( − 4.9 × 10 − 7 and − 0.18 , respectively) and increases DA. On US500, disabling the gate slightly worsens RMSE ( + 0.01 ) and reduces DA. On XAUUSD, the RMSE change is negligible, but DA decreases by 0.15 pp. Overall, noise gating functions as a conservative regularizer whose impact is regime-dependent.
Figure 3. Ablation effects on RMSE and DA. Percentage change in RMSE and DA when each RNAD component is removed from the full objective across four representative assets. Directional supervision has the most consistent effect, while tail gating and noise gating produce smaller, asset-dependent changes.
Figure 3. Ablation effects on RMSE and DA. Percentage change in RMSE and DA when each RNAD component is removed from the full objective across four representative assets. Directional supervision has the most consistent effect, while tail gating and noise gating produce smaller, asset-dependent changes.
Preprints 242159 g003

8. Discussion

This section interprets the empirical findings rather than introducing new experiments. It explains why RNAD improves both RMSE and directional accuracy, how the magnitude and directional terms interact, and why the gains vary across backbones and assets. The goal is to connect the observed benchmark behavior back to the optimization geometry described earlier.

8.1. Why RNAD Improves Forecasting

The consistent RMSE improvements on transformer-style backbones are best understood through gradient signal quality. Standard MSE generates parameter updates proportional to raw residual magnitude, conflating forecast residual with volatility scale. In high-volatility regimes, large shocks dominate gradient updates, steering optimization toward minimizing absolute magnitude rather than learning regime-relative structure. In low-volatility periods, directional noise weakens sign gradients. RNAD addresses both pathologies: volatility normalization calibrates gradient magnitude to predictive difficulty, tail gating bounds the influence of extreme events, and gated directional supervision injects sign information that MSE cannot express.
This interpretation is consistent with a higher signal-to-noise ratio in the gradient stream. By normalizing residuals by s, the loss makes gradient magnitudes reflect how unexpected a residual is relative to the prevailing regime rather than how large it is in absolute terms. The directional term then adds asymmetric penalties around the sign boundary, explicitly encoding the economically motivated half-plane constraint that MSE ignores.

8.2. Directional–Magnitude Coupling

An important empirical finding from the ablation study (Table 10) is that removing the directional penalty (NO_DIR) not only reduces directional accuracy but also consistently increases RMSE. This coupling arises from the geometry of the forecast residual: a sign error amplifies the residual magnitude even when the predicted price level is otherwise close to the true level.

Algebraic illustration.

Let r t = y t − y t − 1 be the realised return and r ^ t = y ^ t − y t − 1 the predicted return. The residual can be written as
e t = y ^ t − y t = r ^ t − r t .
When the signs of r ^ t and r t disagree, the absolute residual | e t | is strictly larger than the maximum of | r t | and | r ^ t | ; in the worst case, where | r ^ t | ≈ | r t | but the signs are opposite, we obtain
e t ≈ − 2 r t ⟹ | e t | ≈ 2 | r t | , and e t 2 ≈ 4 r t 2 .
Thus, a single sign-error event contributes roughly four times as much to the global MSE as a correct-sign prediction of the same magnitude, even if the price forecast is accurate in absolute terms. By penalising directional mistakes, RNAD explicitly pushes r ^ t into the correct half-plane, thereby reducing the frequency of these squared-error amplifying events and directly lowering RMSE—even with a modest directional weight λ dir = 0.5 .

Empirical verification on BTCUSD.

To verify this coupling, we examine the test-set predictions of RNAD on BTCUSD (the most volatile asset, where the effect is most pronounced). Figure 4 displays the distribution of | e t | separated by whether the predicted return sign matched the realised sign. The wrong-sign distribution is markedly shifted toward larger absolute residuals, with a long right tail populated by precisely the type of compounded errors described in Eq. (21). Table 11 reports the corresponding group-level statistics: wrong-sign predictions exhibit an average | e t | that is approximately 35% larger than the correct-sign average (326.89 vs. 242.80). This direct empirical decomposition confirms that sign errors are a primary structural source of large residuals, and that improving directional consistency mechanically reduces RMSE.
In summary, the directional penalty in RNAD acts as an implicit constraint that steers predictions away from the sign-error region, thereby stabilising the magnitude component and producing the joint improvement in RMSE and DA observed throughout the benchmark.

8.3. Architectural Interactions

The architecture-dependent benefit pattern—large gains on transformer-family backbones and smaller gains on N-HiTS—supports the interpretation that loss-level and architecture-level regime awareness partially overlap. N-HiTS’s hierarchical decomposition already captures some regime structure; the marginal benefit of RNAD is correspondingly smaller when the backbone has partially addressed the same failure mode. Importantly, RNAD does not degrade backbone-level aggregate RMSE: on N-HiTS the gain is modest but positive ( Δ RMSE = − 0.086 versus MSE), confirming that the loss is safely composable with regime-aware architectures.

8.4. Statistical Stability

The small standard deviations of RNAD RMSE across seeds ( σ ≤ 0.07 on transformer backbones; Table 7) confirm that the observed improvements are stable across random initializations. This stability is consistent with the gradient-level explanation above: the gains arise from systematic changes in the optimization geometry rather than favorable sampling of the loss landscape.

9. Limitations

Theoretical foundations. 

RNAD is designed on geometric and empirical grounds. No convergence guarantees, Bayes-optimality certificates, or formal proof that the gated directional term constitutes a proper scoring rule (Gneiting & Raftery, 2007) are provided. The EMA volatility estimator introduces first-order autocorrelation in the effective loss landscape and may misrepresent long-memory volatility dynamics not captured by the GARCH-like EMA kernel. Formal Diebold–Mariano or Model Confidence Set tests (Diebold & Mariano, 1995; Hansen et al., 2011) were not applied; extending the benchmark with such comparisons is an important direction for future work.

Hyperparameter sensitivity. 

The loss introduces ten scalar controls (Table 5). While the default values are empirically robust to ± 20 % perturbations for most parameters, τ noise is the most sensitive control and may merit retuning when transferring to a substantially different market universe. The tail-gating parameters τ mag and τ temp are moderately sensitive on heavy-tailed assets (see the BTCUSD ablation results in Table 10) and are the most plausible targets for asset-specific calibration.

Scope of demonstrated improvements. 

Consistent gains are concentrated in one-step-ahead, raw-price prediction on attention-based architectures. Benefits are weaker or absent on regime-aware architectures, at longer forecasting horizons, and in return-domain or log-price formulations where normalization is less critical. Practitioners should validate RNAD on their own setup before assuming transferability.

Economic validity of directional accuracy. 

The DA metric (Eq. 19) weights all non-zero returns equally regardless of magnitude or economic significance, and does not account for transaction costs, bid–ask spread, slippage, position sizing, or the asymmetric utility of large versus small directional errors. Higher DA does not imply higher risk-adjusted returns, and this study makes no such claim. Evaluation within a full trading simulation with realistic costs is necessary to draw conclusions about practical financial value.

Statefulness and engineering overhead. 

The EMA buffer must be synchronized across data-parallel workers, persisted at checkpoints, and reset at each training initialization. This engineering overhead is absent in stateless losses and may complicate deployment in distributed training pipelines.

Batch dependence. 

Because the volatility scale and gates are computed from mini-batches, the objective is mildly batch-size dependent; smaller batches yield noisier scale estimates and can alter the effective gate values. Batch size is held fixed at 64 in all experiments, which controls but does not eliminate this effect.

Evaluation breadth. 

The benchmark uses three seeds and a single chronological split per asset. Extending to multiple non-overlapping temporal windows and applying bootstrap inference over test-set samples would substantially strengthen the statistical conclusions.

10. Conclusion

We introduced RNAD, a regime-normalized adaptive directional loss for financial time-series forecasting. By combining adaptive volatility normalization, tail-robust magnitude handling, and gated directional learning within a single differentiable objective, RNAD aligns the training geometry with the distinctive statistical structure of financial markets: regime-switching, heavy-tailed, and sign-sensitive.
Comprehensive benchmarks across nine assets and four architectures demonstrate consistent RMSE improvements on transformer-style backbones and the lowest per-asset RMSE under PatchTST for every asset. Controlled ablation studies indicate that all three components contribute independently: directional learning is the primary driver of magnitude improvement; tail gating is most critical on heavy-tailed assets; and noise gating contributes modest, regime-dependent regularization. The loss is computationally lightweight, architecture-agnostic, and requires no modification to model structure, data pipelines, or inference code.
The broader implication is that loss-level design is an important lever for financial forecasting improvement, with effects in this benchmark that are of a comparable order of magnitude to several architectural choices while remaining less costly to deploy. Directional accuracy should be interpreted as a forecast-quality metric rather than a direct proxy for trading profitability; this study does not evaluate transaction costs, slippage, or profit and loss. When the forecasting target is volatile, heavy-tailed, and direction-sensitive—as financial prices characteristically are—the training objective can be highly consequential relative to the choice of model architecture.

Data and Code Availability

The implementation code and configuration files required to reproduce the experiments in this manuscript are publicly available at the following GitHub repository: https://github.com/NabeelAhmadSaidd/RNAD Processed data artifacts are not publicly available because of proprietary or ethical restrictions. They are available from the corresponding author on reasonable request and subject to a signed data use agreement.

Conflicts of Interest

The author declares that there are no conflicts of interest, financial or otherwise, that could have influenced the research, authorship, or publication of this work.

A. Supplementary Diagnostics

This appendix provides supporting analyses that complement the main benchmark tables. It includes visualizations of the loss landscape and gradients, volatility-stratified performance, seed stability summaries, a fuller ablation view, and a brief note on computational overhead. These diagnostics are intended to reinforce the main narrative without interrupting the flow of the main text.

A.1. Loss Landscape and Gradient Dynamics

Figure 5 contrasts the gradient vector fields of MSE, Huber, GMADL, and RNAD as functions of normalized error u t and realized return z t . MSE produces gradients that grow linearly with error magnitude and therefore systematically over-weight tail observations in parameter updates. Huber retains MSE-like quadratic behavior near the origin but transitions to linear tails, bounding gradient norms for large residuals. GMADL and RNAD introduce smooth nonlinear damping in the tails: GMADL provides a robust, smoothly saturated gradient response, while RNAD maintains steep, informative gradients near the origin—facilitating precise level learning—and transitions to sub-linear damping during extreme events. Across the four losses, the comparison separates unbounded quadratic sensitivity (MSE), piecewise linear tail control (Huber), and smooth nonlinear tail damping (GMADL and RNAD).
Figure 5. Gradient-field comparison of MSE, Huber, GMADL, and RNAD. MSE gradients grow linearly without bound; Huber uses quadratic small-error and linear tail gradients; GMADL and RNAD apply smooth nonlinear damping to extreme residuals while preserving informative gradients near small errors.
Figure 5. Gradient-field comparison of MSE, Huber, GMADL, and RNAD. MSE gradients grow linearly without bound; Huber uses quadratic small-error and linear tail gradients; GMADL and RNAD apply smooth nonlinear damping to extreme residuals while preserving informative gradients near small errors.
Preprints 242159 g005

A.2. Regime-Stratified Performance

Figure 6 stratifies test performance by causal volatility regime (terciles of realized volatility computed from training-set statistics, to avoid look-ahead bias). RNAD maintains a consistent performance advantage over MSE across all three regimes, indicating that adaptive normalization generalizes across volatility levels rather than simply exploiting tail events.
Figure 6. Performance across volatility regimes. RMSE and DA are shown for BTCUSD, US500, and XAUUSD under low-, medium-, and high-volatility terciles. RNAD retains its advantage over MSE across regimes, indicating that the gain is not confined to a single volatility band.
Figure 6. Performance across volatility regimes. RMSE and DA are shown for BTCUSD, US500, and XAUUSD under low-, medium-, and high-volatility terciles. RNAD retains its advantage over MSE across regimes, indicating that the gain is not confined to a single volatility band.
Preprints 242159 g006

A.3. Seed Stability

Figure 7 shows box-and-whisker distributions of RMSE and DA across nine assets and three seeds. RNAD distributions are consistently shifted toward lower error, with narrow whiskers indicating that improvements arise from the loss geometry rather than favorable initialization.
Figure 7. Seed stability across nine assets. The boxplots summarize PatchTST test RMSE on a log scale and directional accuracy over three random seeds for each loss, showing that RNAD shifts performance toward lower error and higher sign consistency.
Figure 7. Seed stability across nine assets. The boxplots summarize PatchTST test RMSE on a log scale and directional accuracy over three random seeds for each loss, showing that RNAD shifts performance toward lower error and higher sign consistency.
Preprints 242159 g007

A.4. Full Ablation: All Assets and Components

Figure 8 and Figure 9 extend the four-asset ablation reported in Section 7.
Figure 8. Per-asset RMSE change under ablation. The plot shows the change in RMSE relative to the full RNAD loss for each ablation variant on the reported assets, indicating that removing any component is consistently non-beneficial.
Figure 8. Per-asset RMSE change under ablation. The plot shows the change in RMSE relative to the full RNAD loss for each ablation variant on the reported assets, indicating that removing any component is consistently non-beneficial.
Preprints 242159 g008
Figure 9. Component contribution heatmap. Mean RMSE increases from ablating each RNAD component are shown by asset and component, with darker cells indicating greater reliance on that component. The directional term is the most consistent contributor, while tail gating matters most for the most volatile assets.
Figure 9. Component contribution heatmap. Mean RMSE increases from ablating each RNAD component are shown by asset and component, with darker cells indicating greater reliance on that component. The directional term is the most consistent contributor, while tail gating matters most for the most volatile assets.
Preprints 242159 g009

A.5. Computational Overhead

RNAD introduces negligible runtime cost relative to MSE. The EMA update (Eq. 5) requires a single scalar multiply-add per gradient step. Additional loss computations—tanh, logistic gate, softplus—are elementwise operations parallelized by the autograd engine with no synchronization barrier. Memory overhead amounts to one floating-point scalar per asset. In wall-clock benchmarks on all four backbones, training and inference throughput are indistinguishable from MSE at standard batch sizes.
Table 12. Per-epoch training time and peak GPU memory for RNAD compared with MSE, measured on BTCUSD (seed 178, batch size 64) under the exact benchmark configuration. The total wall-clock time for the full 50-epoch training budget is approximated by multiplying the epoch time by 50 (validation overhead is negligible).
Table 12. Per-epoch training time and peak GPU memory for RNAD compared with MSE, measured on BTCUSD (seed 178, batch size 64) under the exact benchmark configuration. The total wall-clock time for the full 50-epoch training budget is approximated by multiplying the epoch time by 50 (validation overhead is negligible).
Backbone Loss Time per epoch (s) Peak GPU memory (MiB) Est. total time (h)
PatchTST MSE 12.17 63.3 0.169
RNAD 13.38 63.3 0.186
iTransformer MSE 11.47 50.4 0.159
RNAD 12.16 50.4 0.169
N-HiTS MSE 8.20 26.7 0.114
RNAD 9.03 26.7 0.125
TimeXer MSE 12.93 34.4 0.180
RNAD 13.91 34.4 0.193
Note: The RNAD overhead consists of a few element-wise operations (EMA update, tanh, sigmoids, softplus) fully parallelised on the GPU, resulting in a negligible increase in runtime and no measurable increase in peak memory.

References

  1. Barron, J. T. (2019). A general and adaptive robust loss function. In Ieee/cvf conference on computer vision and pattern recognition (cvpr) (pp. 4331–4339). [CrossRef]
  2. Bollerslev, T. (1986). Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics, 31(3), 307–327. [CrossRef]
  3. Challu, C., Olivares, K. G., Oreshkin, B. N., Garza, F., Mergenthaler-Canseco, M., & Dubrawski, A. (2023). N-HiTS: Neural hierarchical interpolation for time series forecasting. In Proceedings of aaai (Vol. 37, pp. 6989–6997). [CrossRef]
  4. Charbonnier, P., Blanc-Féraud, L., Aubert, G., & Barlaud, M. (1997). Deterministic edge-preserving regularization in computed imaging. IEEE Transactions on Image Processing, 6(2), 298–311. [CrossRef]
  5. Clark, T. E., & West, K. D. (2007). Approximately normal tests for equal predictive accuracy in nested models. Journal of Econometrics, 138(1), 291–311. [CrossRef]
  6. Cont, R. (2001). Empirical properties of asset returns: Stylized facts and statistical issues. Quantitative Finance, 1(2), 223–236. [CrossRef]
  7. Diebold, F. X., & Mariano, R. S. (1995). Comparing predictive accuracy. Journal of Business & Economic Statistics, 13(3), 253–263. [CrossRef]
  8. Engle, R. F. (1982). Autoregressive conditional heteroskedasticity with estimates of the variance of United Kingdom inflation. Econometrica, 50(4), 987–1007. [CrossRef]
  9. Fan, W., Wang, S., Wang, Y., Wang, X., Zhao, H., Liu, T., Shen, C., Xu, Q., Liu, G., & Wang, X. (2023). Dish-TS: A general paradigm for alleviating distribution shift in time series forecasting. In Proceedings of aaai (Vol. 37, pp. 7522–7529). [CrossRef]
  10. Gneiting, T., & Raftery, A. E. (2007). Strictly proper scoring rules, prediction, and estimation. Journal of the American Statistical Association, 102(477), 359–378. [CrossRef]
  11. Hamilton, J. D. (1989). A new approach to the economic analysis of nonstationary time series and the business cycle. Econometrica, 57(2), 357–384. [CrossRef]
  12. Hansen, P. R., Lunde, A., & Nason, J. M. (2011). The model confidence set. Econometrica, 79(2), 453–497. [CrossRef]
  13. Huber, P. J. (1964). Robust estimation of a location parameter. The Annals of Mathematical Statistics, 35(1), 73–101. [CrossRef]
  14. Kendall, A., Gal, Y., & Cipolla, R. (2018). Multi-task learning using uncertainty to weigh losses for scene geometry and semantics. In Ieee/cvf conference on computer vision and pattern recognition (cvpr) (pp. 7482–7491). [CrossRef]
  15. Kim, T., Kim, J., Tae, Y., Park, C., Choi, J.-H., & Choo, J. (2022). Reversible Instance Normalization for Accurate Time-Series Forecasting against Distribution Shift. In The tenth international conference on learning representations (iclr).
  16. Koenker, R., & Bassett, G. (1978). Regression quantiles. Econometrica, 46(1), 33–50. [CrossRef]
  17. Leitch, G., & Tanner, J. E. (1991). Economic forecast evaluation: Profits versus the conventional error measures. American Economic Review, 81(3), 580–590.
  18. Liu, Y., Hu, T., Zhang, H., Wu, H., Wang, S., Ma, L., & Long, M. (2024). iTransformer: Inverted Transformers Are Effective for Time Series Forecasting. In The twelfth international conference on learning representations (iclr). [CrossRef]
  19. Makridakis, S., Spiliotis, E., & Assimakopoulos, V. (2022). M5 accuracy competition: Results, findings, and conclusions. International Journal of Forecasting, 38(4), 1346–1364. [CrossRef]
  20. Mandelbrot, B. B. (1963). The variation of certain speculative prices. The Journal of Business, 36(4), 394–419. [CrossRef]
  21. Michańków, J., Sakowski, P., & Ślepaczuk, R. (2024). Generalized mean absolute directional loss as a solution to overfitting and high transaction costs in machine learning models used in high-frequency algorithmic investment strategies. (arXiv preprint arXiv:2412.18405). [CrossRef]
  22. Nie, Y., Nguyen, N. H., Sinthong, P., & Kalagnanam, J. (2023). A Time Series is Worth 64 Words: Long-term Forecasting with Transformers. In The eleventh international conference on learning representations (iclr). [CrossRef]
  23. Oreshkin, B. N., Carpov, D., Chapados, N., & Bengio, Y. (2020). N-BEATS: Neural Basis Expansion Analysis for Interpretable Time Series Forecasting. In The eighth international conference on learning representations (iclr). [CrossRef]
  24. Pesaran, M. H., & Timmermann, A. (1992). A simple nonparametric test of predictive performance. Journal of Business & Economic Statistics, 10(4), 461–465. [CrossRef]
  25. Saidd, N. A. (2026a). A controlled comparison of deep learning architectures for multi-horizon financial forecasting: Evidence from 918 experiments. [CrossRef]
  26. Saidd, N. A. (2026b). RADIAN: Regime-Adaptive Dilated Inter-Attention Network for Non-Stationary Financial Time-Series Forecasting. Preprints.org. [CrossRef]
  27. Saleh, R. A., & Saleh, A. K. M. E. (2022). Statistical properties of the log-cosh loss function used in machine learning. [CrossRef]
  28. Salinas, D., Flunkert, V., Gasthaus, J., & Januschowski, T. (2020). DeepAR: Probabilistic forecasting with autoregressive recurrent networks. International Journal of Forecasting, 36(3), 1181–1191. [CrossRef]
  29. van den Oord, A., Dieleman, S., Zen, H., Simonyan, K., Vinyals, O., Graves, A., Kalchbrenner, N., Senior, A., & Kavukcuoglu, K. (2016). Wavenet: A generative model for raw audio. [CrossRef]
  30. Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017). Attention Is All You Need. In Advances in neural information processing systems (Vol. 30, pp. 5998–6008). Curran Associates, Inc. [CrossRef]
  31. Wang, Y., Wu, H., Dong, J., Qin, G., Zhang, H., Liu, Y., Qiu, Y., Wang, J., & Long, M. (2024). Timexer: Empowering transformers for time series forecasting with exogenous variables. [CrossRef]
  32. Zeng, A., Chen, M., Zhang, L., & Xu, Q. (2023). Are transformers effective for time series forecasting? In Proceedings of aaai (Vol. 37, pp. 11121–11128). [CrossRef]
Figure 1. Schematic of the RNAD loss architecture. The diagram illustrates EMA-based volatility estimation and normalization, followed by two parallel loss branches. The tail-adaptive magnitude branch smoothly blends an L2 penalty with a Charbonnier-style robust penalty according to the normalized error magnitude. The noise-aware directional branch uses a smooth gate based on the true normalized change to weight the directional penalty. The two components are combined into the total objective, with the directional loss scaled by λ dir . The EMA state is updated during training and remains unchanged during evaluation.
Figure 1. Schematic of the RNAD loss architecture. The diagram illustrates EMA-based volatility estimation and normalization, followed by two parallel loss branches. The tail-adaptive magnitude branch smoothly blends an L2 penalty with a Charbonnier-style robust penalty according to the normalized error magnitude. The noise-aware directional branch uses a smooth gate based on the true normalized change to weight the directional penalty. The two components are combined into the total objective, with the directional loss scaled by λ dir . The EMA state is updated during training and remains unchanged during evaluation.
Preprints 242159 g001
Figure 2. RNAD loss surface. The surface combines a quadratic basin for small errors, robust tail flattening for large residuals, and directional penalties when predicted and realized signs disagree.
Figure 2. RNAD loss surface. The surface combines a quadratic basin for small errors, robust tail flattening for large residuals, and directional penalties when predicted and realized signs disagree.
Preprints 242159 g002
Figure 4. Distribution of absolute residuals by sign correctness (BTCUSD, test set, RNAD). The density plot separates predictions into correct-sign (green) and wrong-sign (red) groups. The wrong-sign distribution is systematically shifted toward larger | e t | and displays a heavy right tail, providing direct empirical evidence for the directional–magnitude coupling mechanism analysed in Eq. (21).
Figure 4. Distribution of absolute residuals by sign correctness (BTCUSD, test set, RNAD). The density plot separates predictions into correct-sign (green) and wrong-sign (red) groups. The wrong-sign distribution is systematically shifted toward larger | e t | and displays a heavy right tail, providing direct empirical evidence for the directional–magnitude coupling mechanism analysed in Eq. (21).
Preprints 242159 g004
Table 1. Summary of the nine assets used in the benchmark. All series share the same OHLCV-plus-indicator feature set, a close-price target, and an 80 / 10 / 10 chronological split; the volatility proxy is the training-set RMS return used to initialize the EMA state.
Table 1. Summary of the nine assets used in the benchmark. All series share the same OHLCV-plus-indicator feature set, a close-price target, and an 80 / 10 / 10 chronological split; the volatility proxy is the training-set RMS return used to initialize the EMA state.
Asset Market Type Total Train Val Test Vol. Proxy
AAPL Equity 15,673 12,538 1,567 1,568 0.00616
AUDUSD Forex 68,535 54,828 6,853 6,854 0.00127
BTCUSD Crypto 62,793 50,234 6,279 6,280 0.00653
DAX30 Equity Index 57,783 46,226 5,778 5,779 0.00233
ETHUSD Crypto 62,783 50,226 6,278 6,279 0.00847
FTSE100 Equity Index 56,767 45,413 5,676 5,678 0.00197
US500 Equity Index 57,665 46,132 5,766 5,767 0.00200
USDJPY Forex 68,536 54,828 6,853 6,855 0.00096
XAUUSD Commodity 65,093 52,074 6,509 6,510 0.00164
Table 2. Configuration summary for the four forecasting backbones. The table lists the main architecture settings and the retained implementation options used in the benchmark.
Table 2. Configuration summary for the four forecasting backbones. The table lists the main architecture settings and the retained implementation options used in the benchmark.
Backbone Exact configuration Additional settings
PatchTST 3 encoder layers; 4 attention heads; d model = 128 ; d ff = 256 ; patch length = 16 ; stride = 8 ; dropout = 0.05 ; fc-dropout = 0.05 ; head-dropout = 0.0 RevIN enabled; series decomposition disabled; BatchNorm backbone; GELU activation; individual=false; res_attention=true; pe=zeros; learn_pe=true
iTransformer 3 encoder layers; 8 attention heads; d model = 128 ; d ff = 256 ; dropout = 0.05 ; GELU activation Inverted linear embedding over the 24-step input window; LayerNorm encoder; attention factor = 5 ; target index = 3
N-HiTS 3 stacks; 2 blocks per stack; 2 MLP layers per block; n hidden = 256 ; dropout = 0.05 Max pooling; kernel sizes [ 2 , 2 , 2 ] ; downsampling rates [ 4 , 2 , 1 ] ; linear interpolation; ReLU activation; batch normalization disabled; shared weights disabled
TimeXer 2 encoder layers; 8 attention heads; d model = 128 ; d ff = 256 ; patch length = 8 ; dropout = 0.05 ; attention factor = 5 Three non-overlapping target patches over L = 24 ; global token enabled; use_norm=true; GELU activation
Table 6. Main benchmark comparison across four backbones, five losses, nine assets, and three seeds ( { 178 , 956 , 1045 } ). Lower RMSE and MAE are better, higher DA (%) is better, and Δ RMSE reports RNAD’s improvement over the strongest non-RNAD baseline within each backbone. Bold and underlining mark the best and second-best values.
Table 6. Main benchmark comparison across four backbones, five losses, nine assets, and three seeds ( { 178 , 956 , 1045 } ). Lower RMSE and MAE are better, higher DA (%) is better, and Δ RMSE reports RNAD’s improvement over the strongest non-RNAD baseline within each backbone. Bold and underlining mark the best and second-best values.
Backbone Loss RMSE MAE DA (%) Δ RMSE
PatchTST MSE 60.78 ± 0.02 40.36 ± 0.05 50.05 ± 0.20 + 0.004
Huber 60.78 ± 0.02 40.36 ± 0.05 50.02 ± 0.23 + 0.004
log-cosh 60.77 ± 0.02 ̲ 40.35 ± 0.04 50.09 ± 0.19 0.000
GMADL 60.82 ± 0.14 40.27 ± 0.17 ̲ 50.33 ± 0.01 ̲ + 0.051
RNAD 59 . 89 ± 0 . 03 39 . 35 ± 0 . 05 50 . 97 ± 0 . 14 − 0 . 882
iTransformer MSE 61.22 ± 0.32 40.62 ± 0.15 50.28 ± 0.39 + 0.003
Huber 61.23 ± 0.31 40.61 ± 0.19 50.32 ± 0.36 + 0.005
log-cosh 61.22 ± 0.31 ̲ 40.61 ± 0.18 ̲ 50.22 ± 0.06 0.000
GMADL 61.67 ± 0.41 40.82 ± 0.12 50.51 ± 0.49 ̲ + 0.445
RNAD 60 . 54 ± 0 . 07 39 . 84 ± 0 . 02 51 . 29 ± 0 . 10 − 0 . 681
N-HiTS MSE 60.20 ± 0.05 ̲ 39.90 ± 0.13 49.69 ± 0.38 0.000
Huber 60.47 ± 0.38 39.85 ± 0.40 ̲ 50.06 ± 0.39 ̲ + 0.272
log-cosh 60.40 ± 0.29 39.89 ± 0.36 49.83 ± 0.55 + 0.202
GMADL 64.45 ± 4.77 42.39 ± 2.84 49.88 ± 0.37 + 4.256
RNAD 60 . 11 ± 0 . 48 39 . 82 ± 0 . 42 51 . 35 ± 0 . 30 − 0 . 086
TimeXer MSE 60.80 ± 0.18 40.21 ± 0.20 50.23 ± 0.11 + 0.029
Huber 60.78 ± 0.08 40.20 ± 0.13 50.19 ± 0.26 + 0.009
log-cosh 60.84 ± 0.11 40.22 ± 0.14 50.38 ± 0.12 ̲ + 0.069
GMADL 60.77 ± 0.05 ̲ 40.19 ± 0.08 ̲ 50.36 ± 0.07 0.000
RNAD 60 . 12 ± 0 . 01 39 . 58 ± 0 . 02 50 . 81 ± 0 . 24 − 0 . 658
Table 7. Backbone-level averages of RMSE and DA across nine assets and three seeds. The Δ columns report RNAD’s RMSE improvement relative to MSE and GMADL, and the final column identifies the best loss for each backbone.
Table 7. Backbone-level averages of RMSE and DA across nine assets and three seeds. The Δ columns report RNAD’s RMSE improvement relative to MSE and GMADL, and the final column identifies the best loss for each backbone.
Backbone Avg RMSE Avg DA Δ vs. MSE Δ vs. GMADL Best Loss
PatchTST 60 . 61 50.29 − 0 . 886 − 0.933 RNAD
iTransformer 61.18 50 . 52 − 0.684 − 1.126 ̲ RNAD
N-HiTS 61.08 50.16 − 0.086 − 4 . 131 RNAD
TimeXer 60.66 ̲ 50.39 ̲ − 0.687 ̲ − 0.658 RNAD
Table 9. Multi-step ( H = 4 ) forecasting performance of PatchTST with RNAD and MSE on US500 and USDJPY (seed 178).
Table 9. Multi-step ( H = 4 ) forecasting performance of PatchTST with RNAD and MSE on US500 and USDJPY (seed 178).
Dataset Loss RMSE per step DA per step (%) RMSEflat DAavg (%)
H = 1 H = 2 H = 3 H = 4 H = 1 H = 2 H = 3 H = 4
US500 RNAD 14.341 19.966 24.431 28.315 51.27 51.68 52.82 52.32 22.377 52.02
MSE 14.504 20.088 24.459 28.343 49.53 49.36 50.11 49.53 22.447 49.63
USDJPY RNAD 0.180 0.256 0.315 0.367 50.38 51.45 51.92 50.45 0.288 51.05
MSE 0.184 0.258 0.317 0.368 49.43 48.91 50.54 49.67 0.290 49.64
Note: RMSE values are given on the raw price scale and rounded to three decimal places; DA (directional accuracy, %) is rounded to two decimal places. RMSEflat is computed over the flattened 4 × N test samples; DAavg is the arithmetic mean of the four per-step directional accuracies.
Table 11. Average absolute residual | e t | grouped by directional correctness for RNAD on BTCUSD (test set). The wrong-sign group displays a substantially larger mean, consistent with the algebraic doubling effect discussed in the text.
Table 11. Average absolute residual | e t | grouped by directional correctness for RNAD on BTCUSD (test set). The wrong-sign group displays a substantially larger mean, consistent with the algebraic doubling effect discussed in the text.
Prediction Type Count Mean | e t |
Correct-Sign Predictions 3162 242.80
Wrong-Sign Predictions 3088 326.89
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.