Monitoring of near real-time price movement is necessary for data-driven decision making in opening and closing positions for day traders and scalpers. This can be done effectively by constructing a movement path based on forecast distribution of stock prices. High frequency trading data are generally noisy, nonlinear and nonstationary in nature. We develop a quantile recurrent forecasting algorithm via the recurrent algorithm of singular spectrum analysis that can be implemented for any type of time series data. When applied to median forecasting of deterministic and shortand long-memory processes, our quantile recurrent forecast overlaps the true signal. By estimating only the signal dimension number of parameters, this method can construct a recurrent formula by including many lag periods. We apply this method to obtain median forecasts for Facebook, Microsoft, and SNAP’s intraday and daily closing prices. Both for intraday and daily closing prices, the quantile recurrent forecasts produce lower mean absolute deviation from original prices compared to bootstrap median forecasts. We also demonstrate the tracing of price movement over forecast distribution that can be used to monitor stock prices for trading strategy development.

中文翻译：

---

股票价格监测的奇异谱分析中的分位数循环预测

近乎实时的价格变动监控对于日内交易者和剥头皮交易者在开仓和平仓时的数据驱动决策是必要的。这可以通过构建基于股票价格预测分布的移动路径来有效地完成。高频交易数据通常具有噪声、非线性和非平稳性。我们通过奇异谱分析的循环算法开发了一种分位数循环预测算法，该算法可以针对任何类型的时间序列数据实施。当应用于确定性和短期和长期记忆过程的中值预测时，我们的分位数循环预测与真实信号重叠。通过仅估计参数的信号维数，该方法可以构建包含许多滞后周期的递归公式。我们应用这种方法来获得 Facebook、微软和 SNAP 的盘中和每日收盘价的中值预测。对于盘中和每日收盘价，与自举中值预测相比，分位数经常性预测与原始价格的平均绝对偏差更低。我们还演示了跟踪价格变动预测分布，可用于监控股票价格以制定交易策略。