The digitalization of financial services has generated large volumes of borrower data, enabling the use of advanced analytics in credit risk modeling. This study evaluates the performance of statistical and machine learning models for predicting and updating final Loss Given Default (LGD) estimates during recovery, using proprietary data from Norwegian retail credit card portfolios. Traditional regression models (Ordinary Least Squares, Ridge, and Lasso) are compared with ensemble-based machine learning techniques, including Random Forest and Extreme Gradient Boosting (XGBoost). The results show that machine learning models generally achieve lower prediction error, with XGBoost delivering the strongest predictive performance. However, linear models retain greater interpretability and higher explained variance. The findings highlight trade-offs between predictive accuracy and transparency and provide new empirical evidence on dynamic LGD modeling in unsecured retail credit markets.
Bitcoin’s price follows a well-documented power law over long time scales, yet this trend alone is a poor short-term predictor: at horizons of a few months, today’s price outperforms any trend extrapolation. We construct a deliberately minimal baseline, the Naive–Power Law blend (NvPL), which interpolates smoothly from today’s price (at short horizons) to the fitted power-law trend (at long horizons), and ask whether more elaborate models can beat it out of sample. We compare a broad set of alternatives: mean-reversion and momentum corrections, shrinkage regressions (ridge, LASSO, and elastic net), tree-based machine learning (gradient boosting and random forests), and 18 Bayesian structural time series specifications, some hand-specified and some discovered by an autonomous AI agent (the autoresearch pattern). Every model is evaluated on five non-overlapping two-year periods spanning distinct market conditions (2016–2026). For every primary benchmark candidate—the blend and the shrinkage and tree learners—all tunable settings, including the baseline’s own, are selected on training data alone; the Bayesian specifications are deliberately granted more information as a diagnostic. None reliably improves on NvPL. Formal forecast-comparison tests (Diebold–Mariano, Clark–West, and the Model Confidence Set, with inference that treats the forecast origin as the observational unit) sharpen the picture for the primary candidates: no correction significantly improves on NvPL, and the confidence set retains NvPL while excluding the pure power law. The blend’s structure carries detectable predictive content over a pure no-change forecast (Clark–West), but strictly out of sample this content is too small to reliably separate the two in realized error at one to six months: persistence dominates, and the naive forecast is not excluded from the pooled confidence set. Return-based metrics agree, and no model demonstrates directional accuracy above the always-up rule once forecast dependence is accounted for. Aggressive in-sample tuning, including the agent’s, reverses sharply out of sample, illustrating how greedy search overfits non-stationary financial data. At one to six months, short-term Bitcoin prices are dominated by persistence, and NvPL is a stringent, reproducible benchmark that more complex models should be required to beat.
Insider trading detection in cryptocurrency markets is challenging due to the absence of ground truth labels and the introduced pseudonymity. Prior work has focused on rule-based methods, as well as investigated lower-volume data, with limited validation. This work proposes an unsupervised anomaly detection framework validated by model stability. Using public on-chain decentralized exchange trading data, an Isolation Forest was employed and compared across stochastic runs. Given the stability validation, the event specificity of the pre-announcement window and feature representations are investigated. This serves as a case study of the ASTER token and its Coinbase exchange listing. Results show the stability of the produced anomaly structure across feature sets differs significantly. Network-based features produced the highest stability. Stability and event specificity were found to represent distinct properties, as the most stable feature representations were not necessarily event specific. The best performing feature representations across both validation dimensions were the combination of network and behavioural-change features. These findings were consistent under fixed and temporally matched baseline designs. The results suggest that unsupervised anomaly detection, coupled with stability-based validation, provides a practical tool for assessing potential insider trading risk, applicable for estimating Web3 project sustainability.
Spot Ether exchange-traded funds (ETFs) gave regulated investors direct access to the second-largest cryptocurrency in July 2024. We ask whether this form of institutionalization changed the speed and state dependence of Ethereum's response to scheduled macroeconomic news. We align 84 U.S. CPI releases from 2019 to March 2026 with hourly prices, lagged on-chain network activity, and daily spot-ETF flows, and estimate intraday event-study regressions in which the inflation-surprise response is conditioned on the network state. The primary outcome is Ethereum's four-hour return relative to Bitcoin, which differences out the dominant common cryptocurrency component; a rolling market-model abnormal return provides an alternative benchmark. Before the launch, network activity does not materially condition relative inflation pricing. In the ETF era the interaction turns negative, with a pooled change of − 1.13 percentage points (HC3 p = 0.026) that survives wild-bootstrap inference, a max-T adjustment across five horizons, leave-one-announcement diagnostics, and a placebo on pre-announcement returns. The same pattern appears in market-model abnormal returns. The change is concentrated in the four-hour window: it is absent in the release hour and dissipates by 24 hours. Within the ETF era, higher lagged ETF-flow intensity further strengthens the negative response to inflation surprises, although this mechanism test rests on only 18 post-launch announcements. Institutionalization therefore appears to reorganize short-horizon price discovery in digital assets rather than to raise unconditional macro sensitivity, a distinction that daily-frequency designs cannot detect.
This paper investigates whether USDT premium, defined as the deviation of the Tether price from its one-dollar peg in basis points, exhibits systematic and asymmetric behavior across different crisis regimes. Using daily data from January 2019 to January 2026 (n = 2,588), the analysis documents systematic asymmetry in premium behavior across five major crisis episodes. TradFi shocks, characterized by banking stress and USD liquidity events, generate positive cumulative abnormal premiums: COVID-19 produced a CAP of + 1,539 basis points at a twenty-day horizon while the SVB collapse produced + 498 basis points, both significant at the one-percent level. Crypto-native shocks produce the opposite pattern: the FTX collapse generated a CAP of -199 basis points and the LUNA-Terra implosion generated − 169 basis points, significant at the one-percent level. OLS regressions with Newey-West standard errors show that premium persistence (AR coefficient 0.836), BTC return (-58.9), and VIX (+ 0.067) are the principal predictors, whereas DXY shows no significant association once the autoregressive term is included. Local Projection impulse responses show that VIX shocks produce persistent positive responses over a twenty-day window, while BTC return shocks produce the mirror-image negative pattern. Rolling OLS shows that the DXY coefficient is time-varying, unstable, and economically negligible since 2021. These findings are consistent with the interpretation of USDT as an internal USD of the crypto ecosystem, partially decoupled from the external dollar cycle. USDT premium may provide a 24/7 observable market indicator that exhibits systematically different behavior across TradFi stress and crypto-native crisis episodes, without reliance on delayed macroeconomic data.
