• 沒有找到結果。

4.2 策略在路徑上

4.2.2 混淆矩陣

此外,我們也做了測試,對於每一種(過程,策略)的組合均抽樣 100 個,並且 與 Monte Carlo 的結果作比較,此外我們定義策略何為有效:夏普值 > 0 的頻率 超過 0.75(同時對 Monte Carlo 和 RGAN 適用。我们也可以調整這個值來控制我 們的 recall 和 precision)。

表格 4.1 表現的是策略 BH 和過程 GBM 的結果。可以看到總體上的準確率是

Monte Carlo

有效 無效

RGAN

有效 72 0

無效 8 20

表格4.1 GBM 和 BH 的混淆矩陣。

GBM-MAC

Monte Carlo

有效 無效

RGAN

有效 32 9

無效 18 41

表格4.2 GBM 和 MAC 的混淆矩陣。

AR(2)-BH

Monte Carlo

有效 無效

RGAN

有效 0 0

無效 0 100

表格4.3 AR(2)和 BH 的混淆矩陣。

AR(2)-MAC

Monte Carlo

有效 無效

RGAN

有效 39 1

無效 1 59

表格4.4 AR(2)和 MAC 的混淆矩陣。

5 總結

本論文從量化交易中極易碰到的回測過擬合問題出發,希望利用 GAN 來學習到 股價的分佈來緩解這個問題。從機器學習的角度,我們研究了 GAN 學習的性質,

包括生成器、鑒別器的架構以及參數數量,以及在緩解過擬合這個任務下 GAN 對於樣本數量的要求。從應用的角度,我們研究 GAN 所學習的分佈是否有助於 減緩回測過擬合。

我們通過實驗得到以下結論。在我們的假設模型以及選取策略之下:(一)

GAN 可以學習到常用的股價模型的一些性質。(二) 樣本數在這個試驗中的重 要性並不是那麼大。(三) 用 scaling 可以緩解一些在生成階段其值域的相關問 題,但是代價是犧牲了變異數的部分統計性質。(四) GAN 所學到的分佈已經 足夠在我們選擇的一些策略之下某種程度上避免了過擬合這個問題,可以幫助我 們更好的篩選策略。

參考文獻

[1] P. Carr and M. Lopez de Prado, “Determining Optimal Trading Rules without Backtesting,” 2014.

[2] I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, and S. Ozair,

“Generative Adversarial Networks,” in NIPS, 2014, pp. 2672–2680.

[3] D. J. Im, C. D. Kim, H. Jiang, and R. Memisevic, “Generating images with recurrent adversarial networks,” 2016.

[4] D. N. Politis and J. P. Romano, “The Stationary Bootstrap,” in Journal of the American Statistical Association, vol. 89, no. 428, 1994, pp. 1303–1313.

[5] Y. Qin, D. Song, H. Chen, W. Cheng, G. Jiang, and G. Cottrell, “A Dual-Stage Attention-Based Recurrent Neural Network for Time Series Prediction,” 2017.

[6] C. W. J. Granger, “Forecasting stock market prices: Lessons for forecasters,” Int. J.

Forecast., vol. 8, no. 1, pp. 3–13, 1992.

[7] G. Gidófalvi, “Using News Articles to Predict Stock Price Movements,” Dep. Comput.

Sci. Eng. Univ. Calif. San Diego, p. 9, 2001.

[8] M. O. Afolabi and O. Olude, “Predicting Stock Prices Using a Hybrid Kohonen Self Organizing Map (SOM),” Proc. 40th Hawaii Int. Conf. Syst. Sci., pp. 1–8, 2007.

[9] T. S. B. Fletcher, “Machine Learning for Financial Market Prediction,” p. 207, 2012.

[10] D. H. Bailey, J. M. Borwein, M. López de Prado, and Q. J. Zhu, “Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance,” Not. AMS, vol. 61, no. 5, pp. 458–471, 2014.

[11] D. H. Bailey and M. Lopez de Prado, “The Deflated Sharpe Ratio: Correcting for

Selection Bias, Backtest Overfitting and Non-Normality,” J. Portf. Manag., vol. 40, no.

5, pp. 94–107, 2014.

[12] M. Lopez de Prado, D. H. Bailey, J. M. Borwein, M. López de Prado, and Q. J. Zhu,

“The Probability of Backtest Overfitting,” J. Comput. Financ., p. Forthcoming, 2013.

[13] C. R. Harvey and Y. Liu, “Backtesting,” SSRN Electron. J., 2015.

[14] C. M. Investopedia, “Moving Averages: Strategies.” [Online]. Available:

https://www.investopedia.com/university/movingaverage/movingaverages4.asp.

[15] Investopedia, “Buy And Hold.” [Online]. Available:

https://www.investopedia.com/terms/b/buyandhold.asp.

[16] L. Breiman, “Bagging predictors,” Mach. Learn., vol. 24, no. 2, pp. 123–140, 1996.

[17] C. Olah, “Calculus on Computational Graphs: Backpropagation,” 2015. [Online].

Available: http://colah.github.io/posts/2015-08-Backprop/.

[18] S. Hochreiter and J. Urgen Schmidhuber, “Long Short-Term Memory,” Neural Comput., vol. 9, no. 8, pp. 1735–1780, 1997.

[19] C. Olah, “Understanding LSTM Networks,” 2015. [Online]. Available:

http://colah.github.io/posts/2015-08-Understanding-LSTMs/.

[20] S. L. Hyland, C. Esteban, and F. Ratsch, “Real-valued (Medical) Time Series Generation with Recurrent Conditional GANs.” 2017.

[21] S. Nowozin, B. Cseke, and R. Tomioka, “f-GAN: Training Generative Neural Samplers using Variational Divergence Minimization,” 2016.

[22] L. Theis, A. van den Oord, and M. Bethge, “A note on the evaluation of generative models,” in ICLR, 2015, pp. 1–10.

[23] P. Bojanowski, A. Joulin, D. Lopez-Paz, and A. Szlam, “Optimizing the Latent Space of Generative Networks,” 2017.

[24] J. Bergstra and Y. Bengio, “Random Search for Hyper-Parameter Optimization,” J.

Mach. Learn. Res., vol. 13, pp. 281–305, 2012.

[25] K. Kawaguchi, L. P. Kaelbling, and Y. Bengio, “Generalization in Deep Learning,”

2017.

相關文件