Method calculate_future_stats

detection_engine.py:293–347

Calculate different stats for future data to show whether the anomalous stocks found were actually better than non-anomalous ones

(self, predictions_with_output_data)

		print("Results stored successfully in", file_name)

	def calculate_future_stats(self, predictions_with_output_data):
		"""
		Calculate different stats for future data to show whether the anomalous stocks found were actually better than non-anomalous ones
		"""
		future_change = []
		anomalous_score = []
		historical_volatilities = []
		future_volatilities = []

		for item in predictions_with_output_data:
			prediction, symbol, historical_price, future_price = item
			future_sum_percentage_change, future_volatility = self.calculate_future_performance(future_price)
			_, _, historical_volatility = self.calculate_recent_volatility(historical_price)

			# Skip for when there is a reverse split, the yfinance package does not handle that well so percentages get weirdly large
			if abs(future_sum_percentage_change) > 250 or self.is_nan(future_sum_percentage_change) == True or self.is_nan(prediction) == True:
				continue

			future_change.append(future_sum_percentage_change)
			anomalous_score.append(prediction)
			future_volatilities.append(future_volatility)
			historical_volatilities.append(historical_volatility)

		# Calculate correlation and stats
		correlation = np.corrcoef(anomalous_score, future_change)[0, 1]
		anomalous_future_changes = np.mean([future_change[x] for x in range(0, len(future_change)) if anomalous_score[x] < 0]) # Anything less than 0 is considered anomalous
		normal_future_changes = np.mean([future_change[x] for x in range(0, len(future_change)) if anomalous_score[x] >= 0])
		anomalous_future_volatilities = np.mean([future_volatilities[x] for x in range(0, len(future_volatilities)) if anomalous_score[x] < 0]) # Anything less than 0 is considered anomalous
		normal_future_volatilities = np.mean([future_volatilities[x] for x in range(0, len(future_volatilities)) if anomalous_score[x] >= 0])
		anomalous_historical_volatilities = np.mean([historical_volatilities[x] for x in range(0, len(historical_volatilities)) if anomalous_score[x] < 0]) # Anything less than 0 is considered anomalous
		normal_historical_volatilities = np.mean([historical_volatilities[x] for x in range(0, len(historical_volatilities)) if anomalous_score[x] >= 0])
		
		print("\n*************** Future Performance ***************")
		print("Correlation between future absolute change vs anomalous score (lower is better, range = (-1, 1)): **%.2f**\nTotal absolute change in future for Anomalous Stocks: **%.3f**\nTotal absolute change in future for Normal Stocks: **%.3f**\nAverage future volatility of Anomalous Stocks: **%.3f**\nAverage future volatility of Normal Stocks: **%.3f**\nHistorical volatility for Anomalous Stocks: **%.3f**\nHistorical volatility for Normal Stocks: **%.3f**\n" % (
								correlation,
								anomalous_future_changes, normal_future_changes, 
								anomalous_future_volatilities, normal_future_volatilities,
								anomalous_historical_volatilities, normal_historical_volatilities))

		# Plot
		FONT_SIZE = 14
		colors = ['#c91414' if anomalous_score[x] < 0 else '#035AA6' for x in range(0, len(anomalous_score))]
		anomalous_vs_normal = np.array([1 if anomalous_score[x] < 0 else 0 for x in range(0, len(anomalous_score))])
		plt.scatter(np.array(anomalous_score)[anomalous_vs_normal == 1], np.array(future_change)[anomalous_vs_normal == 1], marker='v', color = '#c91414')
		plt.scatter(np.array(anomalous_score)[anomalous_vs_normal == 0], np.array(future_change)[anomalous_vs_normal == 0], marker='P', color = '#035AA6')
		plt.axvline(x = 0, linestyle = '--', color = '#848484')
		plt.xlabel("Anomaly Score", fontsize = FONT_SIZE)
		plt.ylabel("Absolute Future Change", fontsize = FONT_SIZE)
		plt.xticks(fontsize = FONT_SIZE)
		plt.yticks(fontsize = FONT_SIZE)
		plt.legend(["Anomalous", "Normal"], fontsize = FONT_SIZE)
		plt.title("Absolute Future Change", fontsize = FONT_SIZE)
		plt.tight_layout()
		plt.grid()
		plt.show()


# Check arguments