WEBVTT 00:01.000 --> 00:06.610 In this center next videos you will learn how to create sliding windows are moving windows to calculate 00:06.610 --> 00:11.290 rolling statistics like simple moving average as slow and finance. 00:11.290 --> 00:17.920 It is tradition to work with rolling statistics that are often also called financial indicators or financial 00:17.920 --> 00:24.190 studies and rolling statistics are actually heavily used in particular and technical chart analysis 00:24.520 --> 00:27.380 but also for performance reporting. 00:27.400 --> 00:31.290 So what do I mean with moving windows and rolling statistics. 00:31.300 --> 00:38.380 Let's assume that we have daily prices for the S&P 500 index for the last 10 years and we want to know 00:38.380 --> 00:39.780 on each and every day. 00:39.790 --> 00:47.290 The mean or the average stock price of the directly preceding 50 days so on each and every day our window 00:47.290 --> 00:55.180 with 50 daily timestamps moves forward by one day covering their respective day and the forty nine directly 00:55.180 --> 00:58.200 preceding days and from the set. 00:58.300 --> 01:04.870 That we can calculate the mean stock price or the median or the max or the min so the statistic is actually 01:04.870 --> 01:06.180 flexible here. 01:06.370 --> 01:09.800 And by doing so we can actually smooth our data. 01:10.030 --> 01:14.840 But we are also able to capture some recent trends in our price data. 01:14.950 --> 01:17.080 So let's start with an example. 01:17.080 --> 01:24.820 This is always the easiest way and we import pandas and met plop lip and we use the Seabourn style and 01:24.820 --> 01:32.450 then we are importing S&P 500 price data from the CSC profile S&P 500 dot see a sphere. 01:32.470 --> 01:41.050 And as always we try to convert the daytime information to a daytime data type with past dates and we 01:41.050 --> 01:43.260 want to have a date time index. 01:43.300 --> 01:47.770 And also here we pass day to the next column so let's do this. 01:47.770 --> 01:56.180 And by doing so we're actually saving our data frame in the variable S&P 500 and 7. 01:56.500 --> 02:04.450 So these are the first five timestamps starting in the year nineteen hundred seventy and onto the year 02:04.480 --> 02:06.090 2000 and 18. 02:06.100 --> 02:07.690 So the end of the year. 02:07.840 --> 02:12.110 And let's also have a look here at the info method. 02:12.280 --> 02:18.610 So we have twelve thousand one hundred and seven timestamps here and a day time index. 02:18.640 --> 02:22.470 Now let's assume that we only want to work with the closing prices. 02:22.630 --> 02:25.330 So we select the column close. 02:25.360 --> 02:32.080 So this would be a pen series and we can transform a panda series where data frame with the method to 02:32.080 --> 02:36.930 frame and by doing so we're actually overwriting here our variable. 02:37.000 --> 02:38.140 Let's do this. 02:38.140 --> 02:39.490 And let's again have a look here. 02:40.090 --> 02:48.930 So these are the first five timestamps with the first five closing prices of the S&P 500 index. 02:48.940 --> 02:56.470 Now we can also create a chart or a plot with the historic prices by using here the plot method and 02:56.470 --> 02:59.820 the use of fixed size of twelve eight and a font size of 15. 03:00.370 --> 03:02.820 And we also want FBA legend. 03:02.830 --> 03:03.840 So let's have a look here. 03:05.190 --> 03:06.150 So this is here. 03:06.150 --> 03:07.770 Our price chart. 03:07.980 --> 03:14.640 And apparently it wasn't too bad idea to invest in the S&P 500 index in the 70s. 03:14.670 --> 03:22.510 So by doing so we would have increased our investment from let's say 100 to about 2000. 03:22.770 --> 03:27.660 300 and this is still the performance without dividends. 03:27.660 --> 03:33.720 So we do not use yet the adjusted clause but only the clothes and let us decide that we only want to 03:33.720 --> 03:41.340 focus on the last 10 years and select only the timestamps from the thirty first of December 2008 and 03:41.340 --> 03:46.400 to the last day of the year 2000 and 18 and be over right. 03:46.410 --> 03:54.490 Actually our variable here and let's assume that for each and every day of the last 10 years you want 03:54.490 --> 03:58.710 to have a rolling window of 10 times stands for 10 days. 03:58.720 --> 04:03.590 So on each and every day you want to look back on the last nine days. 04:03.610 --> 04:05.710 And actually the respective day. 04:06.010 --> 04:13.670 So this is actually then a moving windows with the main window 10 and in Penang we can actually do this 04:13.670 --> 04:15.530 with the rolling method. 04:15.800 --> 04:23.510 So we have our data from S&P 500 and then we will see the rolling method and we can pass the a 10 to 04:23.510 --> 04:27.240 the windows power meter so let's uh have a look here. 04:27.440 --> 04:35.450 And by doing so we're actually creating a new object of the type or the class Panda's car windows rolling. 04:35.450 --> 04:41.690 So in the background for each and every time stamp panels actually aggregates the directly preceding 04:41.690 --> 04:46.300 a 10 timestamp so we have a rolling window of 10. 04:46.520 --> 04:51.810 And that's again have a look here at the first 15 rows of our S&P 500 data frame. 04:51.860 --> 04:59.370 So this might be quite helpful in a few minutes so we are starting here with the very last day in 2008 04:59.430 --> 05:03.680 and then we have the second of January 2009. 05:03.720 --> 05:09.870 Now let's assume that we want to create here a rolling window of 10 and we are doing the say actually 05:09.870 --> 05:16.230 by applying the rolling method and then we have to change another method we have we have to define how 05:16.230 --> 05:19.100 we want to aggregate actually our window. 05:19.140 --> 05:26.430 So we have actually on each time stamp a set of 10 timestamps and we actually have to define which statistic 05:26.460 --> 05:28.020 we want to calculate. 05:28.020 --> 05:34.170 And for example we can change the mean method and then we are actually calculating the simple moving 05:34.170 --> 05:34.790 average. 05:34.800 --> 05:44.530 So let's do this here and we can see here that um our new data frame starts here with the any end values. 05:44.780 --> 05:52.100 And then here on the 14th of January we have actually the very first the simple moving average or the 05:52.100 --> 05:58.660 very first rolling statistic so apparently on the first nine times stamps welfare and a general use. 05:58.810 --> 06:02.810 And this is simply the case so let's have a look at the very first time stamp. 06:02.830 --> 06:07.300 So here we do not have um the 10 preceding time stamps. 06:07.330 --> 06:11.860 So we only have here one time stamp then for the second time stamp. 06:11.860 --> 06:13.770 We only have two time stamps. 06:13.780 --> 06:18.250 Father rolling statistics then for the third time stamp. 06:18.250 --> 06:22.420 We only have three preceding timestamps and data. 06:23.020 --> 06:27.770 So we cannot calculate a rolling statistic with a window of 10. 06:27.850 --> 06:34.920 And then finally we have the 14th of January it's the tenth time stamp and here we have in total 10 06:35.070 --> 06:38.410 timestamps that are immediately preceding. 06:38.580 --> 06:45.150 And here we are able actually to calculate uh the mean stock price for these 10 times stamps and instead 06:45.150 --> 06:48.510 of calculating the mean we can also calculate the median. 06:48.510 --> 06:50.610 So this is no problem. 06:50.610 --> 06:55.230 And of course we could also calculate uh the maximum. 06:55.230 --> 07:03.630 So the highest stock price of the preceding ten days that's the max method also the minimum with the 07:03.630 --> 07:08.770 main method. 07:08.780 --> 07:10.970 So now let's go back to the million 07:15.130 --> 07:22.560 and actually we have seen here for the first nine time stamps that we get any n values because uh there's 07:22.780 --> 07:30.580 another parameter for the rolling method that is the min periods parameter and by default the argument 07:30.580 --> 07:33.310 for the min periods parameter is actually the same mass. 07:33.310 --> 07:36.760 So the argument for the windows power meter. 07:36.760 --> 07:44.910 So if you pass tend to the windows parameter then by default the min periods parameter is also 10 so 07:44.940 --> 07:46.920 let's have a look at the documentation 07:49.950 --> 07:54.350 so we have here the windows parameter and then the min periods parameter. 07:54.360 --> 07:59.150 And it sets the minimum number of observations and window required to have a value. 07:59.160 --> 08:02.190 Otherwise the result is an A. 08:02.190 --> 08:08.790 So alternatively we could also say that we need a minimum number of observations of a five to calculate 08:08.820 --> 08:15.620 our rolling statistics so let's have a look here and then we can see here that only for the first the 08:15.660 --> 08:22.370 fourth time stamps we have here and well use and then for the fifth time stamp we have in total five 08:22.470 --> 08:24.860 immediately preceding timestamps. 08:25.050 --> 08:31.450 And therefore as we set here minimum periods to five then we can calculate here a rolling statistic. 08:32.160 --> 08:39.540 So this was the introduction on moving windows rolling statistics and simple moving averages and then 08:39.540 --> 08:42.040 the next videos we are going deeper into this topic. 08:42.050 --> 08:43.590 So hope to see you there by.