Bitcoin price prediction has attracted hundreds of academic papers and continuous social media debate, yet the field lacks consensus on even basic questions: can any model beat a naive "today's price" baseline at horizons of one to six months? We survey the peer-reviewed landscape, categorize papers by evaluation methodology, and contrast academic findings with informal but substantive discourse on X/Twitter. The picture that emerges is sobering. At short-to-medium horizons, no peer-reviewed study has shown robust superiority over the naive baseline across multiple market regimes. Daily predictability is real but does not extend to hourly or monthly horizons, and may not survive transaction costs. The stock-to-flow model has failed formal out-of-sample testing, and Metcalfe's Law valuations have been challenged as spurious. The Bitcoin price power law, while empirically compelling, has not been subjected to formal distributional tests. Meanwhile, social media practitioners raise valid statistical critiques -- ordinary least squares (OLS) violations, backtest overfitting, spurious regressions -- that the academic literature has not formalized. We identify open research directions and propose concrete methodological standards for future work -- walk-forward evaluation, multi-regime holdout windows, naive baseline comparison, inclusion of zero in hyperparameter grids, and Diebold-Mariano significance testing -- arguing that the field's primary need is not more models but better evaluation.
翻译:比特币价格预测已吸引了数百篇学术论文及持续的社交媒体辩论,然而该领域甚至在基本问题上缺乏共识:任何模型能否在1至6个月的时间范围内击败简单的“今日价格”基准?我们综述了同行评审的研究现状,按评估方法对论文进行分类,并将学术发现与X/Twitter上非正式但实质性的讨论进行对比。呈现的画面令人警醒。在短期至中期范围内,尚无同行评审研究在多个市场周期中展现出对朴素基准的稳健优势。日频可预测性确实存在,但无法延伸至小时或月频区间,且可能无法覆盖交易成本。库存流量模型未能通过正式样本外检验,梅特卡夫定律估值被质疑为伪相关。比特币价格幂律虽在实证上引人注目,但尚未接受正式分布检验。与此同时,社交媒体实践者提出了有效的统计批判——普通最小二乘法假定违背、回溯测试过拟合、伪回归——这些在学术文献中尚未系统化。我们识别了开放研究方向,并为未来工作提出具体方法论标准——滚动窗口评估、多周期留出样本、朴素基准比较、超参数网格包含零点以及Diebold-Mariano显著性检验——主张该领域的首要需求并非更多模型,而是更优的评估方法。