1 00:00:00,000 --> 00:00:14,700 Wha, good evening, badgeman! 2 00:00:14,700 --> 00:00:36,840 All right. 3 00:00:36,840 --> 00:00:42,340 Welcome to the Cannabis Data Science Meetup for March 24th. 4 00:00:42,340 --> 00:00:46,900 Last week we talked about market concentration. 5 00:00:46,900 --> 00:00:56,500 So this week we'll do a slightly more formal look and we'll actually look at measures of 6 00:00:56,500 --> 00:00:59,260 market concentration. 7 00:00:59,260 --> 00:01:09,420 Just to set the stage, we've been looking at Colorado data and just from the total revenue 8 00:01:09,420 --> 00:01:19,940 we have and the total number of distributors, we can just calculate the average revenue 9 00:01:19,940 --> 00:01:27,200 per distributor just to get a gauge of what the market share may look like. 10 00:01:27,200 --> 00:01:33,300 This may not actually be representative since it's just an average, but if we do a plot, 11 00:01:33,300 --> 00:01:41,020 we'll see that medical average market share is relatively constant and retail decreases 12 00:01:41,020 --> 00:01:50,700 from 2014 to 2017, indicating that the Colorado cannabis market is becoming more competitive 13 00:01:50,700 --> 00:01:53,140 over time. 14 00:01:53,140 --> 00:02:00,980 So we'll try to measure the market concentration and see if it changes over time. 15 00:02:00,980 --> 00:02:04,540 And how will we do that? 16 00:02:04,540 --> 00:02:12,580 Well, the Herfindale-Herschman Index is a measure of market concentration. 17 00:02:12,580 --> 00:02:14,060 This is widely accepted. 18 00:02:14,060 --> 00:02:24,300 It's used in legal cases and it's a go-to tool for economists. 19 00:02:24,300 --> 00:02:28,460 The Herfindale-Herschman Index is a simple calculation. 20 00:02:28,460 --> 00:02:32,540 It's simply the square of all market shares. 21 00:02:32,540 --> 00:02:40,980 For example, if there were two ferns in the market and each one had a 50% market share, 22 00:02:40,980 --> 00:02:48,300 think Coca-Cola has 50% market share, Pepsi has 50% market share, then you would have 23 00:02:48,300 --> 00:02:56,860 a Herfindale-Herschman Index of 5,000. 24 00:02:56,860 --> 00:03:08,700 What you'll note is, as there are more and more firms in the market and they each have 25 00:03:08,700 --> 00:03:16,060 a relatively equal market size, HHI will approach zero. 26 00:03:16,060 --> 00:03:28,220 And if there is one firm with 100, so 100 squared, then the HHI would be 10,000. 27 00:03:28,220 --> 00:03:37,380 So the HHI increases as the market becomes less competitive and it decreases as the market 28 00:03:37,380 --> 00:03:41,020 becomes more competitive. 29 00:03:41,020 --> 00:03:43,060 Why does that matter? 30 00:03:43,060 --> 00:03:52,620 Well, the HHI is often used in legal cases to take into consideration if certain policies 31 00:03:52,620 --> 00:03:57,140 increase or decrease market concentration. 32 00:03:57,140 --> 00:04:07,420 For example, HHI's between 1,500 and 2,500 are considered concentrated. 33 00:04:07,420 --> 00:04:13,420 Anything above 2,500, as we saw earlier, the two firms, that would be considered highly 34 00:04:13,420 --> 00:04:14,640 concentrated. 35 00:04:14,640 --> 00:04:23,500 So if a market truly were dominated by two firms with each 50% market share, that would 36 00:04:23,500 --> 00:04:26,620 be considered a highly concentrated market. 37 00:04:26,620 --> 00:04:33,860 It's not typical to see market shares greater than 30%, but the cannabis industry is new, 38 00:04:33,860 --> 00:04:46,740 so there are turbulent markets and market shares, and a lot of variance in market shares. 39 00:04:46,740 --> 00:04:48,860 Good. 40 00:04:48,860 --> 00:04:56,340 Nick, joining us? 41 00:04:56,340 --> 00:05:02,260 Welcome, Nick. 42 00:05:02,260 --> 00:05:21,900 We are discussing, essentially, we're discussing the average market share, and this can be 43 00:05:21,900 --> 00:05:29,220 measured by the Herfindale-Herschman index, which is just a square of market shares. 44 00:05:29,220 --> 00:05:41,340 And the reason this is important is, for example, if there are a policy decision, and it's going 45 00:05:41,340 --> 00:05:51,740 to increase the Herfindale-Herschman index by more than 200 points, that's going to increase 46 00:05:51,740 --> 00:05:53,940 somebody's market power. 47 00:05:53,940 --> 00:06:04,460 So it's just interesting to see how different policy decisions may affect market concentration. 48 00:06:04,460 --> 00:06:11,860 So we're just going to go ahead and jump into that with the data here. 49 00:06:11,860 --> 00:06:20,480 So as always, the first step is look at the data. 50 00:06:20,480 --> 00:06:26,900 So let me go ahead and... 51 00:06:26,900 --> 00:06:33,380 All right. 52 00:06:33,380 --> 00:06:53,980 And so this will be available on the Cannabis Data Science GitHub here in just one second. 53 00:06:53,980 --> 00:06:59,860 So essentially, what we're going to do is we're going to attempt to measure the market 54 00:06:59,860 --> 00:07:08,260 concentration, and we're going to do it in Washington State. 55 00:07:08,260 --> 00:07:13,580 So we have wholesale sales. 56 00:07:13,580 --> 00:07:16,680 We have it in two different formats here. 57 00:07:16,680 --> 00:07:19,180 So you have it in columns. 58 00:07:19,180 --> 00:07:31,140 So we have monthly wholesale sales for these six months, from August of 2020 to January 59 00:07:31,140 --> 00:07:33,540 of 2021. 60 00:07:33,540 --> 00:07:41,340 And we have the firms and their monthly sales. 61 00:07:41,340 --> 00:07:46,920 And a common way that you can look at this is what it's called panel data. 62 00:07:46,920 --> 00:07:53,740 So this is what you would consider panel data, where you have your organization, you have 63 00:07:53,740 --> 00:07:57,100 their sales, and you have their month. 64 00:07:57,100 --> 00:08:20,940 So you'll see, for example, you'll have each organization, and you'll have their sales 65 00:08:20,940 --> 00:08:23,700 by month. 66 00:08:23,700 --> 00:08:26,820 All right. 67 00:08:26,820 --> 00:08:28,460 So we've got our data. 68 00:08:28,460 --> 00:08:34,820 So now let's see if we can't estimate this. 69 00:08:34,820 --> 00:08:50,460 So we're going to jump over to SPDR and go ahead and read in this data. 70 00:08:50,460 --> 00:09:02,980 And I'm going to try working with the panel data to begin with. 71 00:09:02,980 --> 00:09:10,520 All right. 72 00:09:10,520 --> 00:09:14,700 So let's look at our data, as always. 73 00:09:14,700 --> 00:09:20,660 So we have our organization, our sales, and our month. 74 00:09:20,660 --> 00:09:23,180 All right. 75 00:09:23,180 --> 00:09:28,500 So I've broken this down into the various steps we need to do. 76 00:09:28,500 --> 00:09:33,180 And we're going to see if we can't code this up here in the next 15 minutes. 77 00:09:33,180 --> 00:09:41,780 So we need the total sale by month. 78 00:09:41,780 --> 00:10:03,980 So we can get our months. 79 00:10:03,980 --> 00:10:14,900 All right. 80 00:10:14,900 --> 00:10:19,220 So let's just start hacking at this. 81 00:10:19,220 --> 00:10:32,740 So I guess we just need sales by month. 82 00:10:32,740 --> 00:10:38,900 So there's probably more elegant ways to do this, but let's just get it done to begin 83 00:10:38,900 --> 00:10:39,900 with. 84 00:10:39,900 --> 00:10:46,380 So for month and months. 85 00:10:46,380 --> 00:10:53,900 All right. 86 00:10:53,900 --> 00:10:56,180 So we'll just need to do. 87 00:10:56,180 --> 00:10:57,180 Right. 88 00:10:57,180 --> 00:11:06,140 So data.sales. 89 00:11:06,140 --> 00:11:17,740 We don't want all of the data, so we'll just want to. 90 00:11:17,740 --> 00:11:27,980 Create a page where our month is equal to the month. 91 00:11:27,980 --> 00:11:33,100 So let's see if this snippet of code. 92 00:11:33,100 --> 00:11:37,060 All right. 93 00:11:37,060 --> 00:11:50,300 This is actually some. 94 00:11:50,300 --> 00:11:59,100 I forgot to. 95 00:11:59,100 --> 00:12:07,020 Record the. 96 00:12:07,020 --> 00:12:08,020 All right. 97 00:12:08,020 --> 00:12:15,620 So we now have total sales by month. 98 00:12:15,620 --> 00:12:32,220 So now we need to calculate the market share for each wholesaler. 99 00:12:32,220 --> 00:12:34,020 By month. 100 00:12:34,020 --> 00:12:41,420 So. 101 00:12:41,420 --> 00:12:47,380 That will actually be fairly easy. 102 00:12:47,380 --> 00:12:54,020 To do. 103 00:12:54,020 --> 00:12:56,500 Let's just think about how to do that. 104 00:12:56,500 --> 00:13:01,500 So. 105 00:13:01,500 --> 00:13:20,140 Well. 106 00:13:20,140 --> 00:13:24,980 This is probably not going to be the most elegant way to do things, but we'll come back 107 00:13:24,980 --> 00:13:27,940 and we can refactor as need be. 108 00:13:27,940 --> 00:13:34,100 All right, let's just iterate over the. 109 00:13:34,100 --> 00:13:35,500 The data for him. 110 00:13:35,500 --> 00:13:37,700 So let's just. 111 00:13:37,700 --> 00:13:50,900 Create a series so. 112 00:13:50,900 --> 00:14:14,060 We're just going to iterate over the rows and then the market share is simply going 113 00:14:14,060 --> 00:14:16,060 to be. 114 00:14:16,060 --> 00:14:18,100 The sales. 115 00:14:18,100 --> 00:14:22,380 Divided by. 116 00:14:22,380 --> 00:14:27,220 The sales by month. 117 00:14:27,220 --> 00:14:40,660 Of that particular month. 118 00:14:40,660 --> 00:14:48,900 And I wonder. 119 00:14:48,900 --> 00:14:56,460 If we could just assign this to the data itself. 120 00:14:56,460 --> 00:15:04,100 So. 121 00:15:04,100 --> 00:15:22,500 Think this is. 122 00:15:22,500 --> 00:15:35,380 Yeah. 123 00:15:35,380 --> 00:15:40,300 lt. 124 00:15:40,300 --> 00:15:41,300 Right. 125 00:15:41,300 --> 00:15:58,780 It's sort of having a mind space of how to assign a value to. 126 00:15:58,780 --> 00:16:17,780 And. 127 00:16:17,780 --> 00:16:38,780 Okay, let's see if. 128 00:16:38,780 --> 00:16:47,780 So that dirty piece of code, like I said. 129 00:16:47,780 --> 00:16:50,780 Always refactor. 130 00:16:50,780 --> 00:16:58,780 So we've now calculated the market share for each organization. 131 00:16:58,780 --> 00:17:00,780 So, rule number one. 132 00:17:00,780 --> 00:17:03,780 Look at the data. 133 00:17:03,780 --> 00:17:06,780 So. 134 00:17:06,780 --> 00:17:11,780 Just the mean market share is close to 1%. 135 00:17:11,780 --> 00:17:17,780 And just as a sanity check. 136 00:17:17,780 --> 00:17:25,780 In Colorado, the market share was dipping down to 0.1%. 137 00:17:25,780 --> 00:17:31,780 So that's interesting. 138 00:17:31,780 --> 00:17:39,780 All right. And so we can actually now calculate the HHI by month. 139 00:17:39,780 --> 00:17:47,780 So let's go ahead and do that to see how concentrated these markets are. 140 00:17:47,780 --> 00:17:53,780 And I'm going to do that in a similar way as we calculated total sales by month. 141 00:17:53,780 --> 00:18:10,780 So. 142 00:18:10,780 --> 00:18:25,780 We'll just look at the data for that particular month. 143 00:18:40,780 --> 00:18:50,780 All right. And so then we'll want to calculate the HHI. 144 00:18:50,780 --> 00:19:05,780 And so now we'll iterate over all the firms. So for. 145 00:19:05,780 --> 00:19:11,780 And remember the HHI. 146 00:19:11,780 --> 00:19:24,780 We're going to add the square of the market share. 147 00:19:24,780 --> 00:19:49,780 And we'll keep track of that value. 148 00:19:49,780 --> 00:20:00,780 And let's see if this works. 149 00:20:00,780 --> 00:20:07,780 So. 150 00:20:07,780 --> 00:20:24,780 We now have the Herfindale Hirschman index and let's go ahead and see if we can plot this real quick in the last five minutes, because that's not doing us any good right there. 151 00:20:24,780 --> 00:20:27,780 So. 152 00:20:27,780 --> 00:20:40,780 Let's see if we can create a series out of this somehow. 153 00:20:57,780 --> 00:21:25,780 I just wish I could sort this. 154 00:21:25,780 --> 00:21:46,780 Let's see if we can. 155 00:21:46,780 --> 00:21:59,780 So. 156 00:21:59,780 --> 00:22:27,780 So. 157 00:22:27,780 --> 00:22:37,780 Real quick. 158 00:22:37,780 --> 00:22:50,780 You're right. 159 00:22:50,780 --> 00:23:02,780 Oh, you're correct, Charles. So that's incredibly helpful. 160 00:23:02,780 --> 00:23:21,780 And just to polish this off in the last two minutes here, we have now in 30 minutes, we have discussed market share. 161 00:23:21,780 --> 00:23:30,780 We've introduced the Herfindale Hirschman index, which is a common measure of market concentration. 162 00:23:30,780 --> 00:23:46,780 We've discussed what the HHI is and how low values represent competitive markets and high values represent less competitive markets. 163 00:23:46,780 --> 00:24:01,780 And we've noticed that it can be used in the legal system to determine if policy decisions increase or decrease market concentration. 164 00:24:01,780 --> 00:24:30,780 We have then compiled wholesale sales from open Washington state data. And if you need the data source, we've compiled it here from OpenTHC, who gets their data from the open records requests. 165 00:24:30,780 --> 00:24:40,780 We have then read in the data. We calculated sales by month. 166 00:24:40,780 --> 00:24:48,780 We then calculated market share for each wholesaler by month. 167 00:24:48,780 --> 00:25:04,780 We finally calculated the HHI by month and we plotted it so we can visualize it. So this is a complete data science exercise in 30 minutes. 168 00:25:04,780 --> 00:25:32,780 And our takeaway is it looks like the market concentration is relatively constant and perhaps even decreasing, except for this quite noticeable spike in October of 2020, where there was a sharp decrease in market competitiveness. 169 00:25:32,780 --> 00:25:49,780 So I think I need to end it here to let people get to the Cannabis Sciences Virtual Conference, where they'll be talking about laboratory information management systems for testing cannabis. 170 00:25:49,780 --> 00:26:04,780 However, I'm going to pass this off to the group and discuss why we may have seen a rise in market concentration in October. 171 00:26:04,780 --> 00:26:28,780 So just to throw my idea out there, that is harvest month. But I'll leave that there for the group, because I actually have to go ahead and be wrapping this up today. 172 00:26:28,780 --> 00:26:38,780 I've got to get to the Cannabis Science Conference. 173 00:26:38,780 --> 00:26:50,780 Awesome. So sorry that was a short day, but I'll try to do an extra offering sometime next week in the afternoon to make up for lost time. 174 00:26:50,780 --> 00:26:54,780 So thanks for attending today. 175 00:26:54,780 --> 00:26:56,780 Thank you. It was great. 176 00:26:56,780 --> 00:27:05,780 Oh, I'm glad you learned something and I'll post this code so that you can go through it and extend upon it. 177 00:27:05,780 --> 00:27:06,780 Okay. 178 00:27:06,780 --> 00:27:09,780 Awesome, Charles. And Nick, thank you for coming today. 179 00:27:09,780 --> 00:27:26,780 Have a nice day. Bye.