WEBVTT 00:00:00.000 --> 00:00:01.952 align:middle line:90% [SQUEAKING] 00:00:01.952 --> 00:00:03.904 align:middle line:90% [RUSTLING] 00:00:03.904 --> 00:00:05.368 align:middle line:90% [CLICKING] 00:00:05.368 --> 00:00:16.258 align:middle line:90% 00:00:16.258 --> 00:00:18.800 align:middle line:84% RICHARD DE NEUFVILLE: Hi, this is Richard de Neufville again. 00:00:18.800 --> 00:00:22.070 align:middle line:84% And now, we're switching gears to think not just 00:00:22.070 --> 00:00:26.060 align:middle line:84% about the evidence that we don't forecast correctly, 00:00:26.060 --> 00:00:28.550 align:middle line:84% but let's peer underneath the cover a bit, 00:00:28.550 --> 00:00:30.110 align:middle line:84% underneath the hood, if you want, 00:00:30.110 --> 00:00:32.750 align:middle line:84% of this engine that's driving the forecast 00:00:32.750 --> 00:00:35.510 align:middle line:84% to think about what the forecasting process is 00:00:35.510 --> 00:00:38.660 align:middle line:90% and why it has difficulties. 00:00:38.660 --> 00:00:43.460 align:middle line:84% So the natural thing when anybody is creating a forecast 00:00:43.460 --> 00:00:46.940 align:middle line:84% is to look to the past to see what has been happening. 00:00:46.940 --> 00:00:50.730 align:middle line:84% This is, step one is, what's it's field? 00:00:50.730 --> 00:00:54.330 align:middle line:84% So you look to the past, see what's been moving up or down, 00:00:54.330 --> 00:00:57.230 align:middle line:84% whatever the trend is going to be, and you project this ahead 00:00:57.230 --> 00:00:58.100 align:middle line:90% in some ways. 00:00:58.100 --> 00:01:00.470 align:middle line:84% Now, of course, you want to project it ahead 00:01:00.470 --> 00:01:02.570 align:middle line:84% to take into account some adjustments 00:01:02.570 --> 00:01:04.849 align:middle line:84% so that you are projecting ahead, 00:01:04.849 --> 00:01:07.640 align:middle line:84% but you see there are some changes that people are 00:01:07.640 --> 00:01:09.450 align:middle line:90% using more of it or less of it. 00:01:09.450 --> 00:01:13.760 align:middle line:84% So you just don't project ahead in a straight line 00:01:13.760 --> 00:01:15.200 align:middle line:90% or an exponential line. 00:01:15.200 --> 00:01:16.580 align:middle line:90% You adjust in some way. 00:01:16.580 --> 00:01:19.770 align:middle line:90% That's the common activity. 00:01:19.770 --> 00:01:21.870 align:middle line:90% But now, what then happens? 00:01:21.870 --> 00:01:25.310 align:middle line:84% Let's take a look at some example here. 00:01:25.310 --> 00:01:29.720 align:middle line:84% I'm choosing now some data from the forecast 00:01:29.720 --> 00:01:33.210 align:middle line:84% for international travel from Japan. 00:01:33.210 --> 00:01:35.120 align:middle line:90% So this is the data. 00:01:35.120 --> 00:01:38.870 align:middle line:84% And the important thing is here are 10-year forecasts, 00:01:38.870 --> 00:01:40.940 align:middle line:84% which is the kind of thing you need 00:01:40.940 --> 00:01:43.580 align:middle line:84% if you're planning major investments in an airport. 00:01:43.580 --> 00:01:45.350 align:middle line:84% So what is the traffic going to be? 00:01:45.350 --> 00:01:50.030 align:middle line:84% And what you want to see is on the far right-hand side 00:01:50.030 --> 00:01:52.100 align:middle line:90% here that the percent error. 00:01:52.100 --> 00:01:54.980 align:middle line:84% How big was the difference between what 00:01:54.980 --> 00:01:57.860 align:middle line:84% was forecast 10 years earlier at what came out? 00:01:57.860 --> 00:02:02.810 align:middle line:84% And the numbers here are 65, 53, 27, -13. 00:02:02.810 --> 00:02:04.370 align:middle line:90% That's what the brackets mean. 00:02:04.370 --> 00:02:08.840 align:middle line:84% So not too bad in the last year, really terrible 00:02:08.840 --> 00:02:13.680 align:middle line:84% in the first one that this is the kind of error that happens. 00:02:13.680 --> 00:02:16.340 align:middle line:90% So they weren't accurate. 00:02:16.340 --> 00:02:17.870 align:middle line:90% Why weren't they accurate? 00:02:17.870 --> 00:02:24.270 align:middle line:84% Let's look at the evolution of traffic for this situation. 00:02:24.270 --> 00:02:26.810 align:middle line:84% So the thick blue line with all the dots on it, 00:02:26.810 --> 00:02:28.430 align:middle line:90% that's what actually happened. 00:02:28.430 --> 00:02:31.820 align:middle line:84% And the little dots here underneath 00:02:31.820 --> 00:02:35.390 align:middle line:84% and above and at the end on the right-hand side on it 00:02:35.390 --> 00:02:37.590 align:middle line:90% are the projections. 00:02:37.590 --> 00:02:39.770 align:middle line:90% So what do we observe? 00:02:39.770 --> 00:02:42.470 align:middle line:84% When the traffic was increasing slowly, 00:02:42.470 --> 00:02:45.500 align:middle line:84% naturally, you projected low amounts 00:02:45.500 --> 00:02:49.700 align:middle line:84% so that you got to under-project. 00:02:49.700 --> 00:02:52.760 align:middle line:84% OK, so then they said, OK, when the traffic is increasing 00:02:52.760 --> 00:02:55.980 align:middle line:90% quickly, we will project above. 00:02:55.980 --> 00:03:00.710 align:middle line:84% So you get over the forecast so that whenever 00:03:00.710 --> 00:03:05.540 align:middle line:84% there is something that is starting off slowly and then 00:03:05.540 --> 00:03:10.910 align:middle line:84% going faster, you tend to have these kinds of events 00:03:10.910 --> 00:03:13.760 align:middle line:84% happening-- you under-project, over-project. 00:03:13.760 --> 00:03:16.490 align:middle line:84% So unless the thing is really steady, 00:03:16.490 --> 00:03:19.400 align:middle line:84% as, in the case of Japan, it was later on 00:03:19.400 --> 00:03:23.990 align:middle line:84% and it was fairly even, it's only in those circumstances 00:03:23.990 --> 00:03:26.900 align:middle line:84% that you tend to get answers that are right on. 00:03:26.900 --> 00:03:30.230 align:middle line:84% Otherwise, when you're dealing with new situations, 00:03:30.230 --> 00:03:32.120 align:middle line:84% introduction of new technologies, 00:03:32.120 --> 00:03:34.580 align:middle line:84% as was the case for international travel 00:03:34.580 --> 00:03:38.930 align:middle line:84% from Japan in the '70s and '80s and '90s, when 00:03:38.930 --> 00:03:40.910 align:middle line:84% you have an introduction of new technologies, 00:03:40.910 --> 00:03:45.560 align:middle line:84% you basically have this undershoot and then overshoot. 00:03:45.560 --> 00:03:48.650 align:middle line:90% So what's the pattern? 00:03:48.650 --> 00:03:52.190 align:middle line:84% Forecasting projects for the past and the future 00:03:52.190 --> 00:03:57.320 align:middle line:84% is like steering a car by looking in the rearview mirror. 00:03:57.320 --> 00:03:58.610 align:middle line:90% That's what you're doing. 00:03:58.610 --> 00:04:02.610 align:middle line:84% You're looking to the past and imagining the future. 00:04:02.610 --> 00:04:07.630 align:middle line:84% And what would happen to you if you drove a car like that? 00:04:07.630 --> 00:04:09.375 align:middle line:84% If it's low growth, you underestimate. 00:04:09.375 --> 00:04:12.630 align:middle line:84% If it's high growth, you overestimate, and in general, 00:04:12.630 --> 00:04:13.680 align:middle line:90% it's never right. 00:04:13.680 --> 00:04:15.870 align:middle line:84% And in general, if you actually drove 00:04:15.870 --> 00:04:18.370 align:middle line:84% by looking at the rearview mirror, what would you do? 00:04:18.370 --> 00:04:19.720 align:middle line:90% You'd run off the road. 00:04:19.720 --> 00:04:26.130 align:middle line:84% So one of the issues is that just by the natural processes 00:04:26.130 --> 00:04:28.860 align:middle line:84% of what we do, we have to do in the forecasting, that 00:04:28.860 --> 00:04:32.100 align:middle line:84% is, look to the past to somehow project the future, 00:04:32.100 --> 00:04:35.070 align:middle line:84% there is that tendency to run off the road 00:04:35.070 --> 00:04:38.130 align:middle line:84% whether it is under-forecast in some situations, 00:04:38.130 --> 00:04:40.780 align:middle line:84% over-forecast in some other situations 00:04:40.780 --> 00:04:44.310 align:middle line:84% and, again, never quite get it right. 00:04:44.310 --> 00:04:48.470 align:middle line:90% This is the common experience.