WEBVTT 00:00:00.000 --> 00:00:02.339 Okay, let's unpack this foundational truth of 00:00:02.339 --> 00:00:05.419 successful investing. Don't put all your eggs 00:00:05.419 --> 00:00:07.980 in one basket. Right. It sounds like an ancient 00:00:07.980 --> 00:00:10.419 proverb, something rooted in just simple practical 00:00:10.419 --> 00:00:13.259 wisdom. But what if I told you that less than 00:00:13.259 --> 00:00:15.740 a century ago, this simple concept was mathematically 00:00:15.740 --> 00:00:20.500 isolated, rigorously tested, and codified into 00:00:20.500 --> 00:00:23.019 a Nobel Prize winning science? That's exactly 00:00:23.019 --> 00:00:24.980 the journey we're taking today. We are driving 00:00:24.980 --> 00:00:28.579 deep into modern portfolio theory, or MPT, which 00:00:28.579 --> 00:00:30.899 is really the mathematical framework for diversification. 00:00:31.160 --> 00:00:33.340 It's also known more formally as mean variance 00:00:33.340 --> 00:00:35.500 analysis. We basically took that common sense 00:00:35.500 --> 00:00:37.899 idea and gave us the mathematical rules for exactly 00:00:37.899 --> 00:00:40.899 how and why combining assets efficiently reduces 00:00:40.899 --> 00:00:43.689 risk. So our mission for this deep dive is to 00:00:43.689 --> 00:00:46.409 take this huge stack of sources you provided, 00:00:46.530 --> 00:00:48.350 and they are rich with equations, definitions, 00:00:48.750 --> 00:00:51.490 historical context, and really deliver the critical 00:00:51.490 --> 00:00:53.909 insights. We want you to understand not just 00:00:53.909 --> 00:00:56.770 the terminology of MPT, but how it forced every 00:00:56.770 --> 00:01:00.070 serious investor and money manager to stop viewing 00:01:00.070 --> 00:01:02.549 assets in isolation. And start seeing them as 00:01:02.549 --> 00:01:05.489 interactive parts of a greater whole. Exactly. 00:01:05.790 --> 00:01:09.370 A system. The central goal of MPT is deceptively 00:01:09.370 --> 00:01:13.099 simple. to maximize the expected return for a 00:01:13.099 --> 00:01:15.920 given level of risk. It's a structured approach 00:01:15.920 --> 00:01:18.420 to solving that fundamental trade -off every 00:01:18.420 --> 00:01:20.700 investor faces. And here's where it gets really 00:01:20.700 --> 00:01:22.420 interesting, and this is the core nugget, right? 00:01:22.620 --> 00:01:26.019 The Markowitz insight. This is it. Before MPT, 00:01:26.120 --> 00:01:28.159 a portfolio manager might look at a stock and 00:01:28.159 --> 00:01:30.560 say, that stock is risky. The risk was sort of 00:01:30.560 --> 00:01:33.640 intrinsic to the asset itself, but Harry Markowitz 00:01:33.640 --> 00:01:36.310 completely changed that. He did. His crucial 00:01:36.310 --> 00:01:38.909 insight was that an asset's risk shouldn't be 00:01:38.909 --> 00:01:41.790 assessed on its own, but by how it interacts 00:01:41.790 --> 00:01:44.329 with and contributes to the portfolio's overall 00:01:44.329 --> 00:01:46.989 risk and return profile. That shift in focus 00:01:46.989 --> 00:01:49.890 from individual asset risk to portfolio risk, 00:01:49.989 --> 00:01:51.689 that's everything, isn't it? It's the entire 00:01:51.689 --> 00:01:55.290 game. And when MPT quantifies this risk, it specifically 00:01:55.290 --> 00:01:57.750 uses the variance of return or its square root, 00:01:57.810 --> 00:02:00.090 the standard deviation. And why variance? Was 00:02:00.090 --> 00:02:02.489 that just an arbitrary choice? Not at all. It 00:02:02.489 --> 00:02:05.390 was chosen precisely because it's mathematically 00:02:05.390 --> 00:02:08.189 tractable when you're optimizing these huge, 00:02:08.189 --> 00:02:10.969 complex combinations of assets. It's what allows 00:02:10.969 --> 00:02:14.110 for the constrained quadratic optimization you 00:02:14.110 --> 00:02:16.550 need to actually define the efficient set of 00:02:16.550 --> 00:02:20.030 portfolios. So Markovits introduces MPT in his... 00:02:20.939 --> 00:02:24.419 seminal 1952 paper portfolio selection. And it 00:02:24.419 --> 00:02:26.960 leads much later to his Nobel Prize. Right. But 00:02:26.960 --> 00:02:28.860 our sources point to a really interesting historical 00:02:28.860 --> 00:02:34.060 footnote. Was this concept truly born in 1952? 00:02:34.180 --> 00:02:35.819 Not entirely, at least not in the conceptual 00:02:35.819 --> 00:02:38.080 sense. We can see that the mathematical idea 00:02:38.080 --> 00:02:40.060 of mean variance analysis actually showed up 00:02:40.060 --> 00:02:42.819 earlier. When? In a very obscure paper in 1940 00:02:42.819 --> 00:02:45.419 by an Italian mathematician named Bruno De Finetti. 00:02:45.710 --> 00:02:47.509 And it wasn't even about stocks and bonds, was 00:02:47.509 --> 00:02:50.050 it? No, not at all. It was in the context of 00:02:50.050 --> 00:02:52.710 proportional reinsurance, basically. How insurance 00:02:52.710 --> 00:02:54.710 companies should structure their own risks to 00:02:54.710 --> 00:02:57.229 maximize their expected profit while minimizing 00:02:57.229 --> 00:02:59.969 their volatility. So the same math applies. It 00:02:59.969 --> 00:03:01.990 shows that the mathematical relationship Markowitz 00:03:01.990 --> 00:03:04.949 found is fundamental to managing financial uncertainty 00:03:04.949 --> 00:03:09.740 in any form. But De Finetti's work just... It 00:03:09.740 --> 00:03:12.259 remained unknown to the English -speaking economic 00:03:12.259 --> 00:03:15.199 world until much later. So Markovits gets the 00:03:15.199 --> 00:03:17.840 credit. He gets the credit because he built the 00:03:17.840 --> 00:03:20.300 comprehensive, applicable structure for investment, 00:03:20.539 --> 00:03:23.960 making it the bedrock of modern finance. It's 00:03:23.960 --> 00:03:26.280 a great example of how context and timing can 00:03:26.280 --> 00:03:28.180 matter just as much as the brilliance of the 00:03:28.180 --> 00:03:31.050 idea itself. Okay, so let's move into the mechanics 00:03:31.050 --> 00:03:34.270 of it. If MPT is fundamentally a mathematical 00:03:34.270 --> 00:03:36.810 optimization model, we have to understand the 00:03:36.810 --> 00:03:38.930 inputs and the core assumptions that govern the 00:03:38.930 --> 00:03:41.569 whole equation. What's the behavioral bedrock 00:03:41.569 --> 00:03:44.710 this theory rests on? MPT assumes investors are 00:03:44.710 --> 00:03:47.449 risk -averse. And this isn't just a casual observation, 00:03:47.669 --> 00:03:50.030 it's a necessary input for the model to even 00:03:50.030 --> 00:03:52.669 work. OK, so what does risk averse mean in this 00:03:52.669 --> 00:03:55.229 context? It assumes that if you have two portfolios 00:03:55.229 --> 00:03:57.830 that offer the exact same expected return, a 00:03:57.830 --> 00:03:59.710 rational investor will always choose the one 00:03:59.710 --> 00:04:01.389 with the lower volatility, the lower standard 00:04:01.389 --> 00:04:04.969 deviation. Conversely, the only reason an investor 00:04:04.969 --> 00:04:07.990 would ever accept higher volatility is if they're 00:04:07.990 --> 00:04:10.229 being compensated with a proportionally higher 00:04:10.229 --> 00:04:12.349 expected return. But wait a second. Doesn't that 00:04:12.349 --> 00:04:14.490 assumption immediately get complicated by what 00:04:14.490 --> 00:04:18.029 we see in the real world? How so? Well. The rise 00:04:18.029 --> 00:04:20.529 of highly speculative assets, you know, meme 00:04:20.529 --> 00:04:24.089 stocks, certain crypto assets, it often seems 00:04:24.089 --> 00:04:26.610 like the volatility itself is the attraction, 00:04:26.829 --> 00:04:30.569 not the deterrent. That chase for massive, rapid 00:04:30.569 --> 00:04:34.689 outlier gains seems to contradict pure risk aversion. 00:04:34.870 --> 00:04:37.069 It does. And that's one of the many points where 00:04:37.069 --> 00:04:38.990 NPT gets criticized. And we will definitely get 00:04:38.990 --> 00:04:41.410 to that. OK. But for the model to function mathematically, 00:04:41.850 --> 00:04:43.850 it needs a consistent behavioral assumption. 00:04:44.500 --> 00:04:47.220 MPT sort of abstracts away those psychological 00:04:47.220 --> 00:04:49.800 drivers and just assumes the investor is purely 00:04:49.800 --> 00:04:52.160 rational, focusing only on that quantifiable 00:04:52.160 --> 00:04:55.079 tradeoff between mean return and variance. It's 00:04:55.079 --> 00:04:57.199 a simplifying assumption needed for the optimization 00:04:57.199 --> 00:04:59.680 to work. Exactly. So let's talk about the inputs 00:04:59.680 --> 00:05:01.980 then. We need to accurately measure the return 00:05:01.980 --> 00:05:04.019 of an individual asset, which the scientists 00:05:04.019 --> 00:05:06.180 call three. It's defined as the total net return. 00:05:06.439 --> 00:05:09.699 Why the emphasis on net? Because MPT is designed 00:05:09.699 --> 00:05:12.300 to be grounded in the economic reality of the 00:05:12.300 --> 00:05:15.879 investment. It has to reflect realistic net performance, 00:05:16.240 --> 00:05:19.660 including all costs and all income streams, not 00:05:19.660 --> 00:05:21.860 just theoretical price changes. So the universal 00:05:21.860 --> 00:05:24.240 formula is pretty complex. It includes the change 00:05:24.240 --> 00:05:27.000 in market prices, plus distributions, which is 00:05:27.000 --> 00:05:29.839 deals and goaler, plus accrued interest, civilly 00:05:29.839 --> 00:05:32.199 dollars, and then it subtracts all the transaction 00:05:32.199 --> 00:05:34.459 costs for buying and selling. Right. The civvy 00:05:34.459 --> 00:05:36.459 and de la civvy. So we're not just talking about 00:05:36.459 --> 00:05:38.420 broker commissions. We're talking about all the 00:05:38.420 --> 00:05:41.240 subtle drags on performance, like custodial fees 00:05:41.240 --> 00:05:44.759 or expense ratios. Absolutely. If you're optimizing 00:05:44.759 --> 00:05:47.399 for an efficient frontier. You have to use the 00:05:47.399 --> 00:05:49.240 actual return an investor is putting in their 00:05:49.240 --> 00:05:51.240 pocket. And if we look at that distribution component, 00:05:51.519 --> 00:05:54.899 DTLA, it changes based on the asset class. It's 00:05:54.899 --> 00:05:57.480 dividends for stocks. And coupon payments for 00:05:57.480 --> 00:06:00.139 bonds. Exactly. Now, bonds have that unique feature 00:06:00.139 --> 00:06:02.540 you mentioned, the dirty price. Can you explain 00:06:02.540 --> 00:06:05.740 why MPT mandates using the dirty price, which 00:06:05.740 --> 00:06:08.620 is the clean price, plus accrued interest when 00:06:08.620 --> 00:06:11.100 we're calculating our portfolio weights? Yeah, 00:06:11.120 --> 00:06:13.220 this is a really critical nuance for accuracy. 00:06:13.560 --> 00:06:16.069 Let's imagine you buy a bond for a... clean price 00:06:16.069 --> 00:06:20.350 of, say, $1 ,000. Okay. If that bond pays full 00:06:20.350 --> 00:06:23.170 bonds semi -annually and you buy it three months 00:06:23.170 --> 00:06:25.610 after the last payment, you actually owe the 00:06:25.610 --> 00:06:28.149 seller three months' worth of accrued interest. 00:06:28.449 --> 00:06:31.649 Let's say that's $20. So my actual cash outlay 00:06:31.649 --> 00:06:35.629 is $1 ,020. Precisely. Your actual capital outlay, 00:06:35.689 --> 00:06:38.110 the true fair market value of your position, 00:06:38.350 --> 00:06:42.430 is $1 ,020, the dirty price. And if I calculated 00:06:42.430 --> 00:06:45.050 my portfolio weight based only on the $1 ,000 00:06:45.050 --> 00:06:48.050 clean price, I'd be misrepresenting how much 00:06:48.050 --> 00:06:50.129 capital I've actually dedicated to that asset. 00:06:50.290 --> 00:06:53.009 Exactly right. The portfolio weights, the $20 00:06:53.009 --> 00:06:56.009 down, they have to sum to one, representing 100 00:06:56.009 --> 00:06:59.089 % of your capital. By using the dirty price, 00:06:59.410 --> 00:07:01.850 MPT ensures that the capital base is accurate. 00:07:02.279 --> 00:07:04.459 which is essential before you even try to optimize 00:07:04.459 --> 00:07:06.879 the risk contribution of that asset. So it's 00:07:06.879 --> 00:07:09.160 fundamentally a system designed for precision 00:07:09.160 --> 00:07:11.579 accounting before the math even kicks in. That's 00:07:11.579 --> 00:07:13.379 a great way to put it. Okay, so once we have 00:07:13.379 --> 00:07:15.920 those individual expected returns, calculating 00:07:15.920 --> 00:07:19.089 the portfolio's expected return is... Well, that's 00:07:19.089 --> 00:07:21.029 the easy part. Oh, that's just the weighted average. 00:07:21.110 --> 00:07:24.670 If 60 % of your capital is an asset A with an 00:07:24.670 --> 00:07:27.589 expected return of 10 % and 40 % is an asset 00:07:27.589 --> 00:07:30.750 B with an expected return of 5%. Your portfolio's 00:07:30.750 --> 00:07:33.629 expected return is just 60 % of 10 plus 40 % 00:07:33.629 --> 00:07:37.410 of 5, which is 8%. No complex interactions there. 00:07:37.529 --> 00:07:39.870 Nope. Simple. The complexity and really the core 00:07:39.870 --> 00:07:42.629 of the whole theory is in the portfolio risk 00:07:42.629 --> 00:07:45.310 metrics, the variance. This is where we move 00:07:45.310 --> 00:07:47.870 from simple summation to relational dynamics. 00:07:47.910 --> 00:07:51.029 Yes. For a two -asset portfolio, the variance 00:07:51.029 --> 00:07:53.110 calculation suddenly involves three distinct 00:07:53.110 --> 00:07:56.009 terms. You have the variance of asset A, the 00:07:56.009 --> 00:07:58.470 variance of asset B weighted by their allocations, 00:07:58.589 --> 00:08:00.850 and then you have this crucial correlation component, 00:08:01.129 --> 00:08:06.730 2WA, WB, sigma B. And that coefficient, the correlation 00:08:06.730 --> 00:08:09.649 between asset A and asset B, that's the mathematical 00:08:09.649 --> 00:08:11.889 engine of diversification. It's the entire reason 00:08:11.889 --> 00:08:15.269 NPT exists. And when you scale that up to a real 00:08:15.269 --> 00:08:18.709 portfolio of, say, none at all assets, that simple 00:08:18.709 --> 00:08:21.689 three -term formula just explodes. The risk is 00:08:21.689 --> 00:08:24.350 determined by a vast matrix calculation. Right. 00:08:24.589 --> 00:08:26.389 You still have the non -dollar individual variance 00:08:26.389 --> 00:08:29.370 terms, but they are absolutely dwarfed by the 00:08:29.370 --> 00:08:32.029 complexity of the non -dollar end or two pairwise 00:08:32.029 --> 00:08:34.509 covariances. If you have 100 assets, that's, 00:08:34.509 --> 00:08:38.049 what, 4 ,950 covariance terms you have to calculate. 00:08:38.230 --> 00:08:40.570 So a portfolio's risk is exponentially more about 00:08:40.570 --> 00:08:42.470 the relationships between the assets than the 00:08:42.470 --> 00:08:44.929 risks of the assets themselves. That's it. It 00:08:44.929 --> 00:08:47.350 transforms risk management from a study of singular 00:08:47.350 --> 00:08:50.090 data points into a complex network analysis. 00:08:50.429 --> 00:08:53.269 The optimization process is then inherently about 00:08:53.269 --> 00:08:56.090 finding... the optimal weights, the $2 that minimize 00:08:56.090 --> 00:08:58.590 the collective impact of that whole covariance 00:08:58.590 --> 00:09:01.409 matrix, not just minimizing the individual volatilities. 00:09:01.649 --> 00:09:03.509 And that is what unlocks the free lunch. Ah, 00:09:03.610 --> 00:09:06.529 the free lunch. I mean, given that every investment 00:09:06.529 --> 00:09:09.789 book warns there's no such thing, how can MPT 00:09:09.789 --> 00:09:12.850 confidently promise that diversification is fundamentally 00:09:12.850 --> 00:09:16.190 a risk -free benefit? It's a free lunch because 00:09:16.190 --> 00:09:18.590 the mathematical proof for risk reduction holds 00:09:18.590 --> 00:09:21.370 true without you having to sacrifice any expected 00:09:21.370 --> 00:09:24.580 return. Okay. The diversification principle states 00:09:24.580 --> 00:09:27.960 that an investor reduces portfolio risk, the 00:09:27.960 --> 00:09:30.440 Sigma Muth, by combining instruments that are 00:09:30.440 --> 00:09:33.299 not perfectly positively correlated. As long 00:09:33.299 --> 00:09:35.759 as that correlation coefficient is less than 00:09:35.759 --> 00:09:38.539 one, you benefit. The sources really highlight 00:09:38.539 --> 00:09:41.340 that in a diversified portfolio, the variance 00:09:41.340 --> 00:09:43.940 relies more on the covariance than on the individual 00:09:43.940 --> 00:09:46.919 asset variances. Why does covariance dominate 00:09:46.919 --> 00:09:50.519 the risk equation in practice? Because covariance 00:09:50.519 --> 00:09:53.139 is the only component you can actually control 00:09:53.139 --> 00:09:56.299 through selection. You can't change the intrinsic 00:09:56.299 --> 00:09:58.259 volatility of the sigma per node of an asset. 00:09:58.419 --> 00:10:00.460 If you buy a high -tech stock, it's volatile, 00:10:00.659 --> 00:10:03.139 period. You can't change its nature. Right. But 00:10:03.139 --> 00:10:04.919 you can choose a second asset whose movements 00:10:04.919 --> 00:10:07.500 tend to offset the first one. By selecting assets 00:10:07.500 --> 00:10:10.200 with low or negative correlation, you are actively 00:10:10.200 --> 00:10:13.200 manipulating that covariance term to neutralize 00:10:13.200 --> 00:10:15.100 volatility that would otherwise exist in your 00:10:15.100 --> 00:10:17.500 portfolio. Let's break down the three correlation 00:10:17.500 --> 00:10:20.080 scenarios the sources detail, starting with the 00:10:20.080 --> 00:10:22.200 one that gives you zero diversification benefit. 00:10:22.480 --> 00:10:24.919 That's the benchmark for failure. Perfect positive 00:10:24.919 --> 00:10:27.370 correlation. So everything moves in lockstep. 00:10:27.409 --> 00:10:30.210 Perfect lockstep. Every time asset A jumps 2%, 00:10:30.210 --> 00:10:34.110 asset B jumps 2%. Mathematically, your risk reduction 00:10:34.110 --> 00:10:37.029 is zero. The portfolio's standard deviation, 00:10:37.509 --> 00:10:40.190 sigma peel, is simply the weighted average of 00:10:40.190 --> 00:10:42.789 the individual volatilities. You gain no safety 00:10:42.789 --> 00:10:46.440 from mixing them. So owning, say, Exxon and Chevron 00:10:46.440 --> 00:10:48.860 might offer a tiny bit of risk reduction, but 00:10:48.860 --> 00:10:51.539 owning two ETFs that track the exact same index 00:10:51.539 --> 00:10:54.080 offers none. Precisely. You've effectively just 00:10:54.080 --> 00:10:56.559 created a pseudo single asset. OK, so now let's 00:10:56.559 --> 00:10:58.659 jump to the other extreme, the theoretical ideal, 00:10:58.759 --> 00:11:02.559 which proves the concept. Zero correlation to 00:11:02.559 --> 00:11:05.279 it. Right. If the correlation is zero, that massive 00:11:05.279 --> 00:11:07.639 covariance term in the portfolio variance equation 00:11:07.639 --> 00:11:10.559 just disappears. It vanishes entirely. It does. 00:11:10.840 --> 00:11:14.000 And this allows for the maximum elimination of 00:11:14.000 --> 00:11:16.860 what we call idiosyncratic risk. The formula 00:11:16.860 --> 00:11:19.559 simplifies, and the mathematical proof by Markowitz 00:11:19.559 --> 00:11:22.820 showed that as the number of completely uncorrelated 00:11:22.820 --> 00:11:25.299 assets, that's no dollar approaches infinity, 00:11:25.639 --> 00:11:29.019 the portfolio variance approaches zero. Wow. 00:11:29.159 --> 00:11:31.320 This is the theoretical definition of eliminating 00:11:31.320 --> 00:11:34.559 all non -market risk. The idiosyncratic risk 00:11:34.559 --> 00:11:37.299 is just diversified away. That's amazing. So 00:11:37.299 --> 00:11:40.509 in theory, if I could find infinite, investment 00:11:40.509 --> 00:11:42.870 opportunities that were completely unrelated, 00:11:43.149 --> 00:11:46.470 like the price of copper in Chile, the rental 00:11:46.470 --> 00:11:49.269 rate of property in Berlin, and the revenue of 00:11:49.269 --> 00:11:51.990 a single tech startup in Singapore, I could approach 00:11:51.990 --> 00:11:54.490 a riskless return. You could, theoretically. 00:11:54.789 --> 00:11:56.909 But the reality is the third scenario, partial 00:11:56.909 --> 00:11:59.100 correlation. Zero dollars, Roland Blunt. This 00:11:59.100 --> 00:12:00.519 is where we all live. This is the real world. 00:12:00.559 --> 00:12:02.759 Almost all assets have some correlation because 00:12:02.759 --> 00:12:04.700 they're all influenced by the same macroeconomic 00:12:04.700 --> 00:12:07.059 factors, the same systematic risk. And yet this 00:12:07.059 --> 00:12:09.080 partial correlation is still the source of that 00:12:09.080 --> 00:12:11.740 free lunch. Because as long as one is less than 00:12:11.740 --> 00:12:14.720 one, even if it's 0 .9, the resulting portfolio 00:12:14.720 --> 00:12:17.100 standard deviation will always be lower than 00:12:17.100 --> 00:12:18.840 the simple weighted average of the individual 00:12:18.840 --> 00:12:20.899 standard deviations. And that difference is the 00:12:20.899 --> 00:12:23.080 benefit. That difference is the diversification 00:12:23.080 --> 00:12:25.299 benefit. You get the desired expected return, 00:12:25.539 --> 00:12:27.980 but the volatility is less than what you would 00:12:27.980 --> 00:12:30.639 intuitively calculate just by averaging the component 00:12:30.639 --> 00:12:33.360 parts. So it's the difference between investing 00:12:33.360 --> 00:12:35.799 in two things that are highly related, like two 00:12:35.799 --> 00:12:38.360 airline stocks, versus investing in one airline 00:12:38.360 --> 00:12:41.440 and, say, a utilities company. Exactly. Even 00:12:41.440 --> 00:12:43.340 in a recession, the utility company might be 00:12:43.340 --> 00:12:45.259 relatively stable while the airline crashes, 00:12:45.419 --> 00:12:47.679 and that dampens the total shock to your portfolio. 00:12:48.879 --> 00:12:51.059 The key is that the volatility of the combo is 00:12:51.059 --> 00:12:53.919 lower because their peaks and troughs just don't 00:12:53.919 --> 00:12:56.500 align perfectly. It's the interaction that generates 00:12:56.500 --> 00:12:59.620 the value. OK, so this mathematical heavy lifting 00:12:59.620 --> 00:13:02.200 with the covariance matrix, it leads directly 00:13:02.200 --> 00:13:05.779 to this beautiful visualization that really changed 00:13:05.779 --> 00:13:08.980 financial decision making forever. Let's talk 00:13:08.980 --> 00:13:11.620 about mapping the risk return space and identifying 00:13:11.620 --> 00:13:14.000 the efficient frontier. The risk return space 00:13:14.000 --> 00:13:16.450 is basically our investment map. We plot expected 00:13:16.450 --> 00:13:19.129 return on a vertical axis and risk or standard 00:13:19.129 --> 00:13:21.789 deviation on the horizontal axis. Right. And 00:13:21.789 --> 00:13:25.009 every possible combination of risky assets defined 00:13:25.009 --> 00:13:27.429 by minimizing the variance for a specific target 00:13:27.429 --> 00:13:29.970 return generates a unique point on that map. 00:13:30.129 --> 00:13:33.090 And when you plot all of these possible portfolios, 00:13:33.110 --> 00:13:35.529 they form this curved boundary, which is often 00:13:35.529 --> 00:13:38.250 called the Markowitz bullet. Correct. The boundary 00:13:38.250 --> 00:13:40.649 of this region is defined by solving a constrained 00:13:40.649 --> 00:13:44.309 quadratic optimization problem. Minimize the 00:13:44.309 --> 00:13:47.909 portfolio variance, sigma of 22, subject to achieving 00:13:47.909 --> 00:13:51.370 a specified minimum expected return, ERAP dollars. 00:13:51.970 --> 00:13:54.590 And the result is this hyperbolic shape. And 00:13:54.590 --> 00:13:56.409 the upper part of that hyperbolic boundary is 00:13:56.409 --> 00:13:58.690 what we call the efficient frontier. Why is it 00:13:58.690 --> 00:14:00.570 called efficient, and what does that imply for 00:14:00.570 --> 00:14:02.929 all the portfolios that lie below it? A portfolio 00:14:02.929 --> 00:14:05.590 on that frontier is optimally balanced. If you 00:14:05.590 --> 00:14:07.029 are sitting on the frontier, you've achieved 00:14:07.029 --> 00:14:09.370 the lowest possible risk for that level of expected 00:14:09.370 --> 00:14:12.230 return. or looking at it the other way. The maximum 00:14:12.230 --> 00:14:15.730 possible return for that level of risk. Any portfolio 00:14:15.730 --> 00:14:18.490 that plots below the frontier is suboptimal. 00:14:18.570 --> 00:14:21.169 It's inefficient. It means you are taking on 00:14:21.169 --> 00:14:23.809 more risk than you need to for that return. You're 00:14:23.809 --> 00:14:26.009 getting ripped off. Essentially, the theory is 00:14:26.009 --> 00:14:28.309 telling you, hey, you could move laterally to 00:14:28.309 --> 00:14:30.289 the left, reduce your volatility, and keep the 00:14:30.289 --> 00:14:33.049 exact same return. Precisely. You should either 00:14:33.049 --> 00:14:35.470 adjust your allocation to reduce risk or adjust 00:14:35.470 --> 00:14:38.289 it to increase return until you land back on 00:14:38.289 --> 00:14:40.899 that frontier. The frontier is the standard of 00:14:40.899 --> 00:14:43.220 excellence. And there's one specific point on 00:14:43.220 --> 00:14:44.960 that boundary that is mathematically unique, 00:14:45.179 --> 00:14:48.460 the vertex of the hyperbola. That's the global 00:14:48.460 --> 00:14:52.120 minimum variance portfolio, or GMVP. The GMVP 00:14:52.120 --> 00:14:55.080 represents the portfolio with the absolute lowest 00:14:55.080 --> 00:14:58.159 possible risk you can achieve using the available 00:14:58.159 --> 00:15:01.019 risky assets, regardless of its expected return. 00:15:01.279 --> 00:15:04.029 It's the safest blend possible. What's so fascinating 00:15:04.029 --> 00:15:06.169 about the GMVP is that it often involves some 00:15:06.169 --> 00:15:08.529 pretty non -intuitive allocations, right? Yes. 00:15:08.850 --> 00:15:11.669 For a long -only investor, the GMVP might seem 00:15:11.669 --> 00:15:15.029 surprisingly concentrated. But in a full MPT 00:15:15.029 --> 00:15:17.710 model that allows for short selling, achieving 00:15:17.710 --> 00:15:20.549 the theoretical GMVP often requires you to short 00:15:20.549 --> 00:15:22.809 assets that have high individual volatility, 00:15:23.129 --> 00:15:25.690 a high positive correlation with other assets. 00:15:25.909 --> 00:15:27.769 You short them because their main contribution 00:15:27.769 --> 00:15:31.309 to the portfolio is just... Unwanted covariance. 00:15:31.370 --> 00:15:34.649 Exactly. The GNVP is a mathematically determined 00:15:34.649 --> 00:15:38.049 point of maximum safety, and sometimes achieving 00:15:38.049 --> 00:15:41.110 that means taking on these strange counterintuitive 00:15:41.110 --> 00:15:44.279 negative weights. OK, that brings us to a massive 00:15:44.279 --> 00:15:46.360 simplification for the investor, assuming this 00:15:46.360 --> 00:15:48.659 frontier has already been defined. The two mutual 00:15:48.659 --> 00:15:50.960 fund theorem, also called the separation theorem. 00:15:51.159 --> 00:15:53.200 This is a fundamental result. It says that once 00:15:53.200 --> 00:15:55.580 the complex optimization is done, once that frontier 00:15:55.580 --> 00:15:57.899 is all mapped out, you only need two distinct 00:15:57.899 --> 00:16:00.120 efficient portfolios. Let's call them P to one 00:16:00.120 --> 00:16:02.779 or P to two to create any other portfolio on 00:16:02.779 --> 00:16:04.899 that entire efficient frontier. So an investor 00:16:04.899 --> 00:16:07.399 doesn't need to rebuild the entire covariance 00:16:07.399 --> 00:16:09.820 matrix for 100 stocks. No. They just need to 00:16:09.820 --> 00:16:11.700 find the right combination of two preoptimized. 00:16:11.690 --> 00:16:15.230 optimized funds. Exactly. If PP1 and PT2Rs are 00:16:15.230 --> 00:16:18.250 efficient, your target portfolio, P -target, 00:16:18.250 --> 00:16:21.490 is just a linear mix. Alpha P1 plus one alpha 00:16:21.490 --> 00:16:24.429 P22. Those two funds span the entire frontier. 00:16:24.750 --> 00:16:27.610 This is immensely practical for large asset managers. 00:16:27.870 --> 00:16:30.169 They can provide these two funds and pass the 00:16:30.169 --> 00:16:32.169 complex work onto the investor in a simplified 00:16:32.169 --> 00:16:35.149 form. Just choose your preferred blend, alpha. 00:16:35.289 --> 00:16:37.929 So that weighting factor, alpha, determines your 00:16:37.929 --> 00:16:41.220 position. If I want a standard long -only position 00:16:41.220 --> 00:16:43.379 between PDAO and PTAO dollars, alpha dollars 00:16:43.379 --> 00:16:45.899 is between 0 and 1. Right. But the theorem also 00:16:45.899 --> 00:16:47.980 explains how leverage and short -selling work 00:16:47.980 --> 00:16:50.639 in this space. It does. If you want a portfolio 00:16:50.639 --> 00:16:53.059 that sits outside the segment defined by PDAO 00:16:53.059 --> 00:16:55.200 and PTAO dollars, you need weights that fall 00:16:55.200 --> 00:16:57.799 outside that 0 to 1 range. If alpha is greater 00:16:57.799 --> 00:16:59.919 than 1, you're essentially shorting portfolio 00:16:59.919 --> 00:17:02.539 PDAO dollars, giving it a negative weight, and 00:17:02.539 --> 00:17:04.579 using those proceeds to increase your investment 00:17:04.579 --> 00:17:07.230 in PDAO to more than 100 % of your capital. So 00:17:07.230 --> 00:17:09.269 that's how MPT mathematically incorporates leverage. 00:17:09.589 --> 00:17:12.009 It is. And conversely, an alpha of less than 00:17:12.009 --> 00:17:14.809 zero means shorting P $1 and leveraging into 00:17:14.809 --> 00:17:17.829 two tests too. The math finds the optimal combination, 00:17:18.150 --> 00:17:21.089 even if it involves borrowing or shorting. So 00:17:21.089 --> 00:17:23.170 the two mutual fund theorem simplifies the world 00:17:23.170 --> 00:17:26.630 of risky assets, but MTT takes one more giant 00:17:26.630 --> 00:17:29.569 step toward elegance by introducing a hypothetical, 00:17:29.829 --> 00:17:33.970 truly risk -free asset, $4. This step is crucial. 00:17:34.559 --> 00:17:36.579 It's absolutely crucial because it allows us 00:17:36.579 --> 00:17:39.299 to optimize the entire investment universe, not 00:17:39.299 --> 00:17:42.420 just the risky subset. The risk -free asset is 00:17:42.420 --> 00:17:44.680 defined theoretically as paying a deterministic 00:17:44.680 --> 00:17:47.839 rate. It has zero variance and critically zero 00:17:47.839 --> 00:17:50.559 correlation with any risky asset. And the real 00:17:50.559 --> 00:17:52.539 world, what's our best proxy for that? I mean, 00:17:52.539 --> 00:17:54.519 it's not perfect, but... Typically, it's short 00:17:54.519 --> 00:17:56.400 -term government securities like treasury bills. 00:17:56.599 --> 00:17:58.619 Now, they do carry marginal counterparty risks. 00:17:58.740 --> 00:18:00.279 The government theoretically defaulted inflation 00:18:00.279 --> 00:18:03.160 risk, but their volatility over a short horizon 00:18:03.160 --> 00:18:05.660 is negligible. They're the closest we can get 00:18:05.660 --> 00:18:07.700 to a guaranteed return. And when we combine this 00:18:07.700 --> 00:18:10.099 zero -risk asset four -raft wall with the curve... 00:18:10.119 --> 00:18:12.160 of the Markowitz bullet, the entire geometry 00:18:12.160 --> 00:18:15.400 of investment opportunity just transforms. What 00:18:15.400 --> 00:18:17.819 happens to the efficient frontier? It ceases 00:18:17.819 --> 00:18:20.509 to be a curve. It becomes a straight line. which 00:18:20.509 --> 00:18:22.650 we call the capital allocation line, or CAL. 00:18:22.930 --> 00:18:25.569 A straight line. This line represents the new 00:18:25.569 --> 00:18:28.309 superior set of investment choices because you 00:18:28.309 --> 00:18:31.470 can now combine the risk -free asset with any 00:18:31.470 --> 00:18:34.069 risky portfolio. So the CAL is a straight line 00:18:34.069 --> 00:18:36.430 drawn from the 3LF4 intercept on the vertical 00:18:36.430 --> 00:18:38.829 axis, and it has to be tangent to the Markowitz 00:18:38.829 --> 00:18:41.890 bullet. Why is tangency the critical condition 00:18:41.890 --> 00:18:44.369 here? Because the slope of that line represents 00:18:44.369 --> 00:18:46.990 the Sharpe's ratio. Right, the reward -to -risk 00:18:46.990 --> 00:18:49.470 ratio. The ratio of excess return that's returned. 00:18:49.640 --> 00:18:51.759 above $4 to the risk, which is standard deviation. 00:18:52.099 --> 00:18:54.220 A straight line that is tangent to the curve 00:18:54.220 --> 00:18:57.319 has the highest possible slope. Therefore, the 00:18:57.319 --> 00:18:59.759 point where the CAL touches the risky asset bullet 00:18:59.759 --> 00:19:02.599 is the risky tangency portfolio, which is the 00:19:02.599 --> 00:19:05.400 portfolio of risky assets with the highest attainable 00:19:05.400 --> 00:19:08.119 Sharpe ratio. It offers the maximum reward for 00:19:08.119 --> 00:19:09.880 the risk you're taking. OK, let's map out the 00:19:09.880 --> 00:19:12.319 line itself. The vertical intercept is 100 % 00:19:12.319 --> 00:19:15.700 in the risk -free asset, $3. The tangency point, 00:19:15.819 --> 00:19:18.799 dollars, is 100 % invested in the risky tangency. 00:19:18.859 --> 00:19:21.599 portfolio. So how do we interpret the space between 00:19:21.599 --> 00:19:24.140 those two points? The segment between three and 00:19:24.140 --> 00:19:27.220 dollars represents landing portfolios. An investor 00:19:27.220 --> 00:19:29.640 who is moderately risk -averse will hold a blend, 00:19:29.680 --> 00:19:33.200 say, 70 % in Pillars and 30 % in One of I. They 00:19:33.200 --> 00:19:36.140 are, in effect, lending money at the risk -free 00:19:36.140 --> 00:19:39.099 rate. They accept less risk than portfolio dollars 00:19:39.099 --> 00:19:41.519 offers, and so they get a lower expected return. 00:19:41.819 --> 00:19:43.720 Exactly. And for the aggressive investor who 00:19:43.720 --> 00:19:46.420 wants to move beyond portfolio dollars. They 00:19:46.420 --> 00:19:49.160 move along the CAL beyond the tangency point. 00:19:49.380 --> 00:19:52.309 This represents borrowing portfolios. To get 00:19:52.309 --> 00:19:54.450 returns higher than Teller, the investor has 00:19:54.450 --> 00:19:56.529 to leverage Beller by shorting the risk -free 00:19:56.529 --> 00:19:59.650 asset or, in practical terms, borrowing money 00:19:59.650 --> 00:20:01.630 at the rate -free Arlerns and investing those 00:20:01.630 --> 00:20:04.369 borrowed funds plus their original capital entirely 00:20:04.369 --> 00:20:06.609 into the risky portfolio dollars. And since the 00:20:06.609 --> 00:20:09.079 CAL is a straight line. The increase in expected 00:20:09.079 --> 00:20:11.339 return is perfectly linear with the increase 00:20:11.339 --> 00:20:13.839 in risk. It is. This elegance leads directly 00:20:13.839 --> 00:20:15.980 to the one fund theorem. The ultimate simplification. 00:20:16.220 --> 00:20:18.619 It really is. The theorem states that all investors 00:20:18.619 --> 00:20:20.799 can achieve their optimal risk return profile 00:20:20.799 --> 00:20:23.680 using only the risk -free asset and that single 00:20:23.680 --> 00:20:27.079 globally optimal risky fund. The tangency portfolio. 00:20:27.420 --> 00:20:29.920 So regardless of your personal risk tolerance, 00:20:30.240 --> 00:20:32.700 the math says you should hold pausers. You should 00:20:32.700 --> 00:20:34.940 hold pausers. Your risk appetite just determines 00:20:34.940 --> 00:20:37.680 the blend. how much origin you lend or borrow 00:20:37.680 --> 00:20:39.880 relative to your holding of dollars. Now, the 00:20:39.880 --> 00:20:43.039 sources mention a technical caveat, the possibility 00:20:43.039 --> 00:20:46.099 of the tangency portfolio escaping to infinity. 00:20:46.359 --> 00:20:48.519 What does that mean for the model's validity? 00:20:48.900 --> 00:20:51.859 It's a point where the model can fail. The tangency 00:20:51.859 --> 00:20:54.640 portfolio only exists if the risk -free rate 00:20:54.640 --> 00:20:57.059 is less than the return of the global minimum 00:20:57.059 --> 00:21:00.650 variance portfolio. to a main VP. OK. If free 00:21:00.650 --> 00:21:03.269 dollars were to somehow become extremely high, 00:21:03.369 --> 00:21:06.829 higher than the GMVP, the CEL would become parallel 00:21:06.829 --> 00:21:10.130 to the hyperbola's asymptote. In that theoretical 00:21:10.130 --> 00:21:12.470 scenario, the model suggests the optimal portfolio 00:21:12.470 --> 00:21:14.829 would diverge, requiring infinite shorting of 00:21:14.829 --> 00:21:17.130 some assets and infinite longing of others. So 00:21:17.130 --> 00:21:19.150 if the return on risk free T -bills is somehow 00:21:19.150 --> 00:21:21.569 higher than the safest possible combination of 00:21:21.569 --> 00:21:23.829 stocks, the model breaks down because it implies 00:21:23.829 --> 00:21:27.200 an obvious infinite arbitrage opportunity. Exactly. 00:21:27.259 --> 00:21:29.940 It acts as a necessary theoretical check on the 00:21:29.940 --> 00:21:33.000 inputs. It forces us to confirm that the inputs 00:21:33.000 --> 00:21:35.859 are economically sound before the optimization 00:21:35.859 --> 00:21:38.220 can give us a meaningful result. Okay, so this 00:21:38.220 --> 00:21:40.799 entire framework, MPT, the efficient frontier, 00:21:41.079 --> 00:21:43.579 the CAL, it's all the setup for understanding 00:21:43.579 --> 00:21:46.180 how assets must be priced in an efficient market. 00:21:46.640 --> 00:21:49.299 If all rational investors are following the CAL 00:21:49.299 --> 00:21:52.160 and holding that single optimal risky fund, pie 00:21:52.160 --> 00:21:55.319 dollars, what happens when the market reaches 00:21:55.319 --> 00:21:57.940 equilibrium? In a state of market equilibrium, 00:21:58.299 --> 00:22:00.920 everyone is holding the risky assets in identical 00:22:00.920 --> 00:22:03.380 proportions. And those proportions are defined 00:22:03.380 --> 00:22:06.480 by the tangency portfolio. Right. Since the aggregated 00:22:06.480 --> 00:22:08.380 demand from all investors matches the available 00:22:08.380 --> 00:22:11.099 supply of all assets in the market, the tangency 00:22:11.099 --> 00:22:14.140 portfolio becomes the market portfolio. The single 00:22:14.140 --> 00:22:16.579 most efficient risky portfolio is, in fact, the 00:22:16.579 --> 00:22:19.079 market itself. And this foundational link between 00:22:19.079 --> 00:22:21.460 MPT and the market portfolio is what gives us 00:22:21.460 --> 00:22:24.519 the Capital Asset Pricing Model, or CAPM. But 00:22:24.519 --> 00:22:26.400 before we get to the formula, we have to understand 00:22:26.400 --> 00:22:28.480 the core insight about risk compensation, which 00:22:28.480 --> 00:22:32.019 hinges on risk decomposition. Yes. The total 00:22:32.019 --> 00:22:34.700 risk of any individual's security has to be broken 00:22:34.700 --> 00:22:37.460 down into two components, and the market only 00:22:37.460 --> 00:22:40.599 compensates you for one of them. First, there's 00:22:40.599 --> 00:22:43.500 specific risk. Idiosyncratic risk, unique risk, 00:22:43.680 --> 00:22:46.619 diversifiable risk. That's the risk tied to the 00:22:46.619 --> 00:22:49.839 single company, right? The CEO resigns, a product 00:22:49.839 --> 00:22:52.200 fails. Correct. The factory burns down. Because 00:22:52.200 --> 00:22:54.720 these events are unique to the company and uncorrelated 00:22:54.720 --> 00:22:56.740 with the broader economy, they tend to cancel 00:22:56.740 --> 00:22:58.559 each other out when you combine them with thousands 00:22:58.559 --> 00:23:01.660 of other assets in a large portfolio. MPT proves 00:23:01.660 --> 00:23:03.539 this risk is eliminated through diversification. 00:23:03.819 --> 00:23:05.640 And the second component is the one you can't 00:23:05.640 --> 00:23:08.140 get rid of. That's systematic risk, market risk, 00:23:08.420 --> 00:23:10.720 non -diversifiable risk. This risk is driven 00:23:10.720 --> 00:23:13.559 by universal macroeconomic forces. Inflation, 00:23:13.599 --> 00:23:15.880 war, interest rate changes, recession fears. 00:23:16.420 --> 00:23:18.680 Because it affects all assets at the same time, 00:23:18.720 --> 00:23:21.279 you cannot diversify it away. So the core principle 00:23:21.279 --> 00:23:23.960 of compensation that drives CAPM has to be this. 00:23:24.099 --> 00:23:26.720 If I can eliminate specific risk for free by 00:23:26.720 --> 00:23:29.299 simply diversifying, then market has no incentive 00:23:29.299 --> 00:23:31.619 to pay me a premium for bearing it. Absolutely. 00:23:31.759 --> 00:23:34.579 The market is rational and competitive. Therefore, 00:23:34.859 --> 00:23:38.140 the market only provides a risk premium for bearing 00:23:38.140 --> 00:23:41.200 systematic risk. And so an asset's expected return 00:23:41.200 --> 00:23:44.000 isn't determined by its total variance. But only 00:23:44.000 --> 00:23:46.240 by its measure of sensitivity to overall market 00:23:46.240 --> 00:23:48.519 movements. And that sensitivity measure is the 00:23:48.519 --> 00:23:52.460 now famous factor, beta. Beta is the quantitative 00:23:52.460 --> 00:23:55.680 realization of systematic risk. It's the asset 00:23:55.680 --> 00:23:58.900 sensitivity relative to the market. Mathematically, 00:23:59.039 --> 00:24:01.940 it's calculated as the ratio of the asset's covariance 00:24:01.940 --> 00:24:04.779 with the market, that's sigma i, to the market's 00:24:04.779 --> 00:24:07.700 own variance, sigma m into 2. So a beta of 1 00:24:07.700 --> 00:24:10.319 means the asset moves in line with the market. 00:24:10.460 --> 00:24:13.319 A beta of 1 .5 means that if the market moves 00:24:13.319 --> 00:24:16.839 up 10%, the asset is expected to move up 15%. 00:24:16.839 --> 00:24:19.480 And if the market falls 10%, the asset is expected 00:24:19.480 --> 00:24:22.680 to fall 15%. It's a measure of leverage to systematic 00:24:22.680 --> 00:24:25.200 movement. And that beta factor is what determines 00:24:25.200 --> 00:24:28.079 the asset's theoretical required return, which 00:24:28.079 --> 00:24:30.920 is plotted on the security market line, or SML. 00:24:31.140 --> 00:24:33.220 Yes, and the SML states that the relationship 00:24:33.220 --> 00:24:36.160 between systematic risk beta and expected return 00:24:36.160 --> 00:24:39.200 must be linear. If two assets have the same beta, 00:24:39.359 --> 00:24:42.180 they must offer the same expected return, regardless 00:24:42.180 --> 00:24:44.980 of how much individual diversifiable risk they 00:24:44.980 --> 00:24:47.059 carry. Okay, let's define the ultimate quantitative 00:24:47.059 --> 00:24:50.119 realization of this concept, the CAPM equation 00:24:50.119 --> 00:24:52.839 itself. The formula for the theoretical required... 00:24:52.880 --> 00:24:56.500 Expected return, E -ray, is E -ary, E -y -r -f, 00:24:56.599 --> 00:25:00.279 plus beta, E -ary -r -m -r -f. So it breaks the 00:25:00.279 --> 00:25:02.500 required return into two components. It does. 00:25:02.579 --> 00:25:04.900 The time value of money, which is the risk -free 00:25:04.900 --> 00:25:07.279 rate, risk -foidy way, and then the compensation 00:25:07.279 --> 00:25:10.000 for systematic risk. The term E -r -m -r -f -e 00:25:10.000 --> 00:25:12.839 is the market risk premium, the extra return 00:25:12.839 --> 00:25:15.240 the market offers over the risk -free rate. The 00:25:15.240 --> 00:25:17.380 logic here is so powerful. It's really worth 00:25:17.380 --> 00:25:19.259 walking through the logical derivation, even 00:25:19.259 --> 00:25:21.480 without showing all the calculus. Why must this 00:25:21.480 --> 00:25:24.059 formula hold true in equilibrium? It all rests 00:25:24.059 --> 00:25:26.839 on that foundational assumption of MPT, that 00:25:26.839 --> 00:25:29.680 the market is efficient and there can be no uncompensated 00:25:29.680 --> 00:25:32.660 risk. So imagine we have two assets, A and B. 00:25:33.299 --> 00:25:35.539 If we were to add a marginal amounts asset A 00:25:35.539 --> 00:25:38.599 to the market portfolio, more dollar, we would 00:25:38.599 --> 00:25:40.960 gain a marginal expected return and take on a 00:25:40.960 --> 00:25:43.099 marginal amount of risk. And since the market 00:25:43.099 --> 00:25:45.980 is in equilibrium, the ratio of that marginal 00:25:45.980 --> 00:25:48.480 expected return to the marginal risk has to be 00:25:48.480 --> 00:25:51.299 the same for every single asset we consider adding. 00:25:51.519 --> 00:25:53.779 Exactly. If the marginal reward to risk ratio 00:25:53.779 --> 00:25:56.180 for asset A were higher than that for asset B, 00:25:56.319 --> 00:25:59.339 investors would flock to asset A, driving its 00:25:59.339 --> 00:26:02.079 price up and its expected return down until the 00:26:02.079 --> 00:26:04.559 ratio is equalized. So in equilibrium, the marginal 00:26:04.559 --> 00:26:07.000 reward to risk ratio for any asset must equal 00:26:07.000 --> 00:26:09.500 the reward... to risk ratio of the market portfolio 00:26:09.500 --> 00:26:12.259 itself. That's the trainer ratio equality. And 00:26:12.259 --> 00:26:14.480 by using the definitions of return, risk, and 00:26:14.480 --> 00:26:16.640 covariance, and setting the marginal contribution 00:26:16.640 --> 00:26:18.680 of an asset equal to the marginal contribution 00:26:18.680 --> 00:26:21.319 of the market, the algebraic rearrangement just 00:26:21.319 --> 00:26:24.619 forces the equation into the CAPM format. It's 00:26:24.619 --> 00:26:27.000 a self -consistent proof that the required compensation 00:26:27.000 --> 00:26:29.940 is entirely determined by beta. It's an intellectual 00:26:29.940 --> 00:26:32.680 powerhouse. So how is this applied in practice? 00:26:32.759 --> 00:26:35.180 If I calculate this required return, what do 00:26:35.180 --> 00:26:37.849 I do with it? You use it for valuation. The required 00:26:37.849 --> 00:26:40.930 return, yin -yang, acts as the discount rate 00:26:40.930 --> 00:26:43.789 to calculate an asset's intrinsic present value, 00:26:43.890 --> 00:26:46.509 or value dollars, based on its expected future 00:26:46.509 --> 00:26:49.130 cash flows. And if the calculated value is higher 00:26:49.130 --> 00:26:51.990 than the current market price, the asset is theoretically 00:26:51.990 --> 00:26:55.029 undervalued. And it should offer a positive abnormal 00:26:55.029 --> 00:26:57.930 return. And that abnormal return leads us to 00:26:57.930 --> 00:26:59.769 the metric that defines the success of active 00:26:59.769 --> 00:27:02.910 managers, alpha. Alpha is the measure of performance 00:27:02.910 --> 00:27:05.430 above and beyond what the systematic risk or 00:27:05.430 --> 00:27:07.430 beta of the portfolio should have generated. 00:27:07.990 --> 00:27:11.009 Statistically, we estimate CAPM using a regression 00:27:11.009 --> 00:27:13.509 known as the Security Characteristic Line, or 00:27:13.509 --> 00:27:16.589 SCL, which plots the asset's excess return against 00:27:16.589 --> 00:27:19.569 the market's excess return. Alpha is the intercept 00:27:19.569 --> 00:27:22.569 of this regression. In a world where CAPM holds 00:27:22.569 --> 00:27:26.230 perfectly, alpha should be zero, right? Precisely. 00:27:26.230 --> 00:27:29.690 If alpha is positive, the manager has generated 00:27:29.690 --> 00:27:32.869 excess return. through superior security selection 00:27:32.869 --> 00:27:36.089 or timing. In other words, successfully navigating 00:27:36.089 --> 00:27:38.829 the diversifiable idiosyncratic risk portion. 00:27:39.049 --> 00:27:41.630 And that is the active manager's entire goal 00:27:41.630 --> 00:27:44.329 to prove they are skilled enough to generate 00:27:44.329 --> 00:27:47.210 an alpha greater than zero consistently. MPT 00:27:47.210 --> 00:27:50.289 is a beautiful, mathematically sound theory that 00:27:50.289 --> 00:27:53.769 earned the Nobel Prize. And yet it is also arguably 00:27:53.769 --> 00:27:56.490 the most criticized concept in modern finance. 00:27:56.690 --> 00:27:58.710 Let's tackle the fundamental flaws, starting 00:27:58.710 --> 00:28:01.240 with the biggest practical weakness. estimation. 00:28:01.440 --> 00:28:03.880 Yeah, this is a big one. MPT is built entirely 00:28:03.880 --> 00:28:07.059 on expected values, future returns, future volatility, 00:28:07.359 --> 00:28:10.019 future correlations. But those inputs are, by 00:28:10.019 --> 00:28:12.500 definition, unobservable. We can't see the future. 00:28:12.660 --> 00:28:14.500 Right. So in practice, we're forced to substitute 00:28:14.500 --> 00:28:16.660 historical data as proxies for these expectations. 00:28:17.019 --> 00:28:18.859 And the fundamental problem with using history 00:28:18.859 --> 00:28:21.000 to predict the future is that the future often 00:28:21.000 --> 00:28:23.579 contains circumstances that history simply hasn't 00:28:23.579 --> 00:28:26.500 recorded yet. Yeah, exactly. For MPT optimization 00:28:26.500 --> 00:28:29.500 to work optimally, you need an accurate forward 00:28:29.500 --> 00:28:32.680 look. covariance matrix. But the matrix you calculate 00:28:32.680 --> 00:28:36.019 today is based on, say, a last 10 years of data. 00:28:36.160 --> 00:28:38.700 That data knows nothing about the next major 00:28:38.700 --> 00:28:42.000 technological disruption or a sudden pandemic 00:28:42.000 --> 00:28:44.940 or a rapid geopolitical regime change. So MPT 00:28:44.940 --> 00:28:47.799 is intrinsically backward looking, but its results 00:28:47.799 --> 00:28:50.160 are used for forward looking decisions. Yes. 00:28:50.160 --> 00:28:52.960 And that's a huge disconnect. A second. Much 00:28:52.960 --> 00:28:55.759 deeper mathematical criticism targets the bedrock 00:28:55.759 --> 00:28:58.099 assumption that allows the calculations to be 00:28:58.099 --> 00:29:01.000 so elegant in the first place. The Gaussian assumption. 00:29:01.279 --> 00:29:03.579 Right. NPT assumes that asset returns follow 00:29:03.579 --> 00:29:06.299 a Gaussian or normal distribution, the classic 00:29:06.299 --> 00:29:09.039 bell curve. This is what allows standard deviation 00:29:09.039 --> 00:29:11.500 to be a complete measure of risk. But critics, 00:29:11.619 --> 00:29:13.920 including Nobel laureate Eugene Fama and the 00:29:13.920 --> 00:29:16.460 brilliant fractal mathematician Benoit Mandelbrot, 00:29:16.559 --> 00:29:19.559 show this is just inadequate for real world finance. 00:29:19.799 --> 00:29:21.920 It really is. So what's the reality of financial 00:29:21.920 --> 00:29:24.420 returns compared to that idealized bell curve? 00:29:24.799 --> 00:29:27.460 Real world returns are characterized by liptocritosis. 00:29:27.720 --> 00:29:30.680 They have very high peaks and critically fat 00:29:30.680 --> 00:29:32.980 tails. Fat tails. Which means that small changes 00:29:32.980 --> 00:29:35.720 happen frequently. That's the high peak. But 00:29:35.720 --> 00:29:39.140 extreme events, those massive rapid market crashes 00:29:39.140 --> 00:29:42.119 or booms, happen far more frequently than the 00:29:42.119 --> 00:29:44.960 normal distribution predicts. The Gaussian curve. 00:29:45.539 --> 00:29:48.539 severely, severely underestimates the probability 00:29:48.539 --> 00:29:51.400 of these outlier events. So MPT models look at 00:29:51.400 --> 00:29:53.299 historical data, they calculate the standard 00:29:53.299 --> 00:29:55.500 deviation, and then they tell an investor that 00:29:55.500 --> 00:29:58.140 a three standard deviation event, a massive crisis, 00:29:58.420 --> 00:30:00.759 should happen maybe once every thousand years. 00:30:01.019 --> 00:30:02.599 And then it happens three times in a decade. 00:30:02.799 --> 00:30:05.839 Right. Nassim Nicholas Taleb famously criticized 00:30:05.839 --> 00:30:09.319 this in the wake of the 2008 crisis. He argued 00:30:09.319 --> 00:30:11.920 that MPT models, based on the Gaussian assumption, 00:30:12.160 --> 00:30:14.539 were essentially hot air. He said the models 00:30:14.539 --> 00:30:16.809 were beautifully platonic, but fundamentally 00:30:16.809 --> 00:30:19.309 disconnected from reality because they failed 00:30:19.309 --> 00:30:21.650 to properly account for the non -normal frequency 00:30:21.650 --> 00:30:24.309 of these large market movements, what he termed 00:30:24.309 --> 00:30:27.349 black swans. And the failure isn't just in calculating 00:30:27.349 --> 00:30:29.710 the probability of loss, but in calculating the 00:30:29.710 --> 00:30:31.750 correlation spike. What happens to that beautiful 00:30:31.750 --> 00:30:34.589 diversification benefit during a crisis? That's 00:30:34.589 --> 00:30:37.609 the most devastating practical failure. Diversification 00:30:37.609 --> 00:30:40.660 works when droas is less than one. But during 00:30:40.660 --> 00:30:43.640 a systemic crisis like the one that struck LTCM 00:30:43.640 --> 00:30:46.579 in 1998 or the global financial crisis in 2008, 00:30:46.859 --> 00:30:49.640 the one thing that NPT optimization relies on 00:30:49.640 --> 00:30:53.259 completely collapses. Correlation spikes to one. 00:30:53.380 --> 00:30:56.099 All assets start moving in lockstep, usually 00:30:56.099 --> 00:30:59.119 downward. So the diversification benefit vanishes 00:30:59.119 --> 00:31:01.960 at the exact moment you need it most. Absolutely. 00:31:02.160 --> 00:31:04.839 The model tells you the assets are poorly correlated. 00:31:05.079 --> 00:31:07.980 But in the specific structural scenario of a 00:31:07.980 --> 00:31:10.339 liquidity crunch or a mass deleveraging, all 00:31:10.339 --> 00:31:12.579 bets are off and that correlation matrix becomes 00:31:12.579 --> 00:31:14.380 meaningless. Let's talk about the distinction 00:31:14.380 --> 00:31:16.859 between probabilistic versus structural risk. 00:31:17.079 --> 00:31:20.119 This is a major philosophical gap between MPT 00:31:20.119 --> 00:31:22.279 and traditional engineering risk management. 00:31:22.420 --> 00:31:25.579 MPT is purely probabilistic. It deals only with 00:31:25.579 --> 00:31:27.400 numerical likelihoods derived from historical 00:31:27.400 --> 00:31:29.720 frequency. It can tell you there's a 5 % chance 00:31:29.720 --> 00:31:32.140 of a 10 % loss, but it says nothing about the 00:31:32.140 --> 00:31:34.839 underlying causality or interconnectedness. Whereas 00:31:34.839 --> 00:31:37.839 a nuclear engineer using a PRA model looks at 00:31:37.839 --> 00:31:41.900 system relationships. If component A fails, does 00:31:41.900 --> 00:31:44.720 it overload component B? They're modeling the 00:31:44.720 --> 00:31:47.859 structure of the failure. NPT completely lacks 00:31:47.859 --> 00:31:50.539 that structural model. It cannot compute the 00:31:50.539 --> 00:31:53.710 odds of a novel system -level event, say, a sudden 00:31:53.710 --> 00:31:55.690 collapse in confidence driven by social media 00:31:55.690 --> 00:31:58.529 that has no historical precedent because there's 00:31:58.529 --> 00:32:00.730 no historical data to feed the variance calculation. 00:32:01.150 --> 00:32:03.630 The model is blind to novel structural risks 00:32:03.630 --> 00:32:06.029 and feedback loops. And finally, there's a key 00:32:06.029 --> 00:32:09.150 behavioral flaw, the symmetry of variance and 00:32:09.150 --> 00:32:11.569 loss aversion. Yeah, variance is mathematically 00:32:11.569 --> 00:32:14.210 symmetrical. It treats a 10 % positive surprise 00:32:14.210 --> 00:32:18.289 good volatility as equally risky as a 10 % negative 00:32:18.289 --> 00:32:21.079 surprise, which is bad volatility. But we know 00:32:21.079 --> 00:32:23.259 from Kahneman and Tversky about loss aversion. 00:32:23.460 --> 00:32:26.099 The pain of a loss is psychologically two to 00:32:26.099 --> 00:32:27.819 two and a half times greater than the pleasure 00:32:27.819 --> 00:32:30.880 of an equivalent gain. And since MPT treats gains 00:32:30.880 --> 00:32:33.259 and losses symmetrically, its measure of risk 00:32:33.259 --> 00:32:35.839 often doesn't align with an investor's true psychological 00:32:35.839 --> 00:32:38.980 preference. Investors care deeply about downside 00:32:38.980 --> 00:32:42.000 risk, which MPT just bundles together with upside 00:32:42.000 --> 00:32:44.809 potential. And this disconnect led to other theories, 00:32:44.990 --> 00:32:47.910 right? It did. It led to the development of postmodern 00:32:47.910 --> 00:32:51.650 portfolio theory, or PMPT, which uses asymmetric 00:32:51.650 --> 00:32:53.890 risk measures like semi -variance or downside 00:32:53.890 --> 00:32:57.089 deviation. It focuses solely on volatility below 00:32:57.089 --> 00:32:59.569 a target return, trying to capture that true 00:32:59.569 --> 00:33:01.769 sense of loss aversion. And we can't forget the 00:33:01.769 --> 00:33:04.490 traditional critiques from, say, value investors, 00:33:04.789 --> 00:33:08.150 those who focus on intrinsic business value rather 00:33:08.150 --> 00:33:10.910 than statistical inputs. Contrarian investors 00:33:10.910 --> 00:33:13.450 like Sir John Templeton often express profound 00:33:13.450 --> 00:33:15.950 skepticism about MPT's reliance on historical 00:33:15.950 --> 00:33:19.130 data. They argue that focusing on unreliable 00:33:19.130 --> 00:33:21.470 and irrelevant statistical inputs like historical 00:33:21.470 --> 00:33:24.589 volatility just distracts from fundamental analysis. 00:33:25.009 --> 00:33:27.910 For a value investor, risk isn't about volatility. 00:33:27.950 --> 00:33:30.400 It's about paying too much for an asset. or the 00:33:30.400 --> 00:33:33.000 risk of permanent capital loss, which MPT doesn't 00:33:33.000 --> 00:33:36.000 explicitly model at all. So MPT may have its 00:33:36.000 --> 00:33:37.980 flaws, but the underlying concepts, optimization, 00:33:38.420 --> 00:33:40.460 efficiency, measuring return against volatility, 00:33:40.720 --> 00:33:43.099 are so general that they've proven useful in 00:33:43.099 --> 00:33:45.519 some really surprising domains, completely outside 00:33:45.519 --> 00:33:47.740 of stocks and bonds. Let's look at some of these 00:33:47.740 --> 00:33:50.220 real -world use cases. A very common application, 00:33:50.539 --> 00:33:52.880 especially in corporate finance, is optimizing 00:33:52.880 --> 00:33:56.210 project portfolios. A large pharmaceutical company, 00:33:56.289 --> 00:33:58.849 for instance, has a portfolio of new drug development 00:33:58.849 --> 00:34:01.910 projects. Each project has an expected return, 00:34:02.210 --> 00:34:06.289 its profitability, and a defined risk, the chance 00:34:06.289 --> 00:34:08.309 of failure. And the optimization goal is the 00:34:08.309 --> 00:34:11.269 same. Maximize the expected return of the entire 00:34:11.269 --> 00:34:13.909 slate of projects for a given level of aggregate 00:34:13.909 --> 00:34:16.920 risk. But projects are fundamentally different 00:34:16.920 --> 00:34:19.460 from liquid financial assets. You mentioned lumpiness. 00:34:19.820 --> 00:34:22.320 That is the main caveat. Financial assets are 00:34:22.320 --> 00:34:24.900 continuously divisible. You buy one share or 00:34:24.900 --> 00:34:27.679 1 ,000 shares. Projects are often lumpy. They're 00:34:27.679 --> 00:34:30.000 all or nothing. You either fund the new factory, 00:34:30.099 --> 00:34:33.000 100 % committed, or you don't, 0%. Which makes 00:34:33.000 --> 00:34:35.059 the modeling much more complex. Significantly 00:34:35.059 --> 00:34:38.159 more complex than standard MPT itemization, which 00:34:38.159 --> 00:34:40.750 assumes continuous divisibility. And the illiquidity 00:34:40.750 --> 00:34:43.630 factor is huge, too. Absolutely. If a stock starts 00:34:43.630 --> 00:34:46.030 to fail, you can sell it instantly and minimize 00:34:46.030 --> 00:34:48.630 your losses. If you abandon a half -complete 00:34:48.630 --> 00:34:52.090 drug trial, the sunk cost is likely a 100 % loss 00:34:52.090 --> 00:34:54.349 of capital. So how do you measure risk? Well, 00:34:54.409 --> 00:34:56.750 since projects don't have historical variances 00:34:56.750 --> 00:34:59.849 based on daily price movements, the risk is typically 00:34:59.849 --> 00:35:02.670 expressed in more subjective terms, like probability 00:35:02.670 --> 00:35:06.070 of failing to achieve cost of capital or chance 00:35:06.070 --> 00:35:08.690 of losing more than half the investment. But 00:35:08.690 --> 00:35:11.050 the conceptual structure of finding the efficient 00:35:11.050 --> 00:35:14.150 frontier for projects remains valid. Okay, where 00:35:14.150 --> 00:35:16.710 else have these quantitative optimization ideas 00:35:16.710 --> 00:35:19.889 cropped up? Well, in the 1970s, MPT concepts 00:35:19.889 --> 00:35:22.070 were adopted in regional science and economics. 00:35:22.530 --> 00:35:24.570 Researchers started modeling the composition 00:35:24.570 --> 00:35:27.510 of a regional labor force or an industry mix 00:35:27.510 --> 00:35:30.289 using these portfolio theoretic methods. So they're 00:35:30.289 --> 00:35:32.630 trying to optimize the regional economy's portfolio 00:35:32.630 --> 00:35:35.530 of different industries to achieve stable growth. 00:35:35.730 --> 00:35:38.369 Stable growth while minimizing cyclical variability. 00:35:38.929 --> 00:35:41.250 That's fascinating. Treating a city's industrial 00:35:41.250 --> 00:35:44.130 base like a stock portfolio, trying to make sure 00:35:44.130 --> 00:35:46.190 that when the manufacturing sector is down, the 00:35:46.190 --> 00:35:49.039 service sector is up. Exactly. It leverages the 00:35:49.039 --> 00:35:51.780 covariance concept, trying to ensure that different 00:35:51.780 --> 00:35:53.980 segments of the economy are not perfectly correlated, 00:35:54.239 --> 00:35:57.320 which provides economic stability. Even more 00:35:57.320 --> 00:36:00.340 abstractly, MPT has been applied to social psychology 00:36:00.340 --> 00:36:03.239 to model the self -concept. The self -concept. 00:36:03.239 --> 00:36:05.480 How does your sense of self become a portfolio? 00:36:06.159 --> 00:36:08.639 Researchers theorize that the attributes, identities, 00:36:08.900 --> 00:36:11.500 and roles that define a person, parent, employee, 00:36:11.860 --> 00:36:14.739 athlete, friend act as psychological assets. 00:36:15.320 --> 00:36:17.460 The strength and independence of these assets 00:36:17.460 --> 00:36:20.000 determine your psychological stability. Ah, so 00:36:20.000 --> 00:36:22.619 if your entire sense of self -esteem relies only 00:36:22.619 --> 00:36:26.059 on one domain, say your career, and you lose 00:36:26.059 --> 00:36:28.579 your job, your whole psychological portfolio 00:36:28.579 --> 00:36:31.360 suffers massive volatility. That's the idea. 00:36:31.760 --> 00:36:33.820 Diversifying your identity is the psychological 00:36:33.820 --> 00:36:37.380 equivalent of diversification. Reducing the portfolio's 00:36:37.380 --> 00:36:40.019 standard deviation. A well -diversified self 00:36:40.019 --> 00:36:43.139 -concept with low covariance between roles provides 00:36:43.139 --> 00:36:45.320 a kind of built -in psychological insurance against 00:36:45.320 --> 00:36:48.099 failure in any single domain. That's a remarkable 00:36:48.099 --> 00:36:51.800 generalization of the core math. It is. And finally, 00:36:51.920 --> 00:36:54.699 NPT principles are even used in information retrieval. 00:36:54.969 --> 00:36:57.090 When search engines rank documents, they're not 00:36:57.090 --> 00:36:59.070 just looking for maximum relevance, which you 00:36:59.070 --> 00:37:00.929 can think of as expected return. They're also 00:37:00.929 --> 00:37:03.530 trying to minimize what? Minimize the uncertainty 00:37:03.530 --> 00:37:06.570 or ambiguity of the ranked list, the risk. The 00:37:06.570 --> 00:37:09.030 goal is to maximize the overall relevance of 00:37:09.030 --> 00:37:11.030 the set of retrieved documents while minimizing 00:37:11.030 --> 00:37:13.369 the chance that the user clicks on a series of 00:37:13.369 --> 00:37:16.510 irrelevant links. Documents become assets in 00:37:16.510 --> 00:37:19.050 an information portfolio. Hashtag tag tag outro. 00:37:19.579 --> 00:37:22.219 This deep dive really shows that modern portfolio 00:37:22.219 --> 00:37:24.920 theory is an undeniable intellectual achievement. 00:37:25.159 --> 00:37:27.500 It formalized the free lunch of diversification. 00:37:27.860 --> 00:37:30.119 It established the efficient frontier as the 00:37:30.119 --> 00:37:32.639 mathematically optimal boundary. And it fundamentally 00:37:32.639 --> 00:37:34.920 proved that the market only compensates investors 00:37:34.920 --> 00:37:37.219 for bearing systematic risk, which is measured 00:37:37.219 --> 00:37:40.099 by beta. And despite its reliance on statistical 00:37:40.099 --> 00:37:43.320 proxies and that often flawed Gaussian assumption, 00:37:43.460 --> 00:37:46.489 which, you know. led to it underestimating the 00:37:46.489 --> 00:37:48.389 probability of fat -tailed black swan events, 00:37:48.650 --> 00:37:50.889 NPT remains the fundamental building block for 00:37:50.889 --> 00:37:53.670 asset pricing and risk management globally. It 00:37:53.670 --> 00:37:55.409 gives you the necessary language in the map to 00:37:55.409 --> 00:37:57.639 understand how risk compensation works. I think 00:37:57.639 --> 00:37:59.820 the ultimate lesson MPT provides is that risk 00:37:59.820 --> 00:38:02.440 is not inherent to what you own, but rather about 00:38:02.440 --> 00:38:04.800 what you own relative to everything else. So 00:38:04.800 --> 00:38:07.219 given the criticisms that MPT failed to predict 00:38:07.219 --> 00:38:10.239 major structural shifts and systemic crises precisely 00:38:10.239 --> 00:38:12.980 because it lacks a structural model, here is 00:38:12.980 --> 00:38:14.860 our final provocative thought for you to chew 00:38:14.860 --> 00:38:18.199 on. If the next major risk to our global portfolios 00:38:18.199 --> 00:38:20.940 isn't just financial volatility, but rather non 00:38:20.940 --> 00:38:23.840 -financial systemic structures like rapid climate 00:38:23.840 --> 00:38:26.780 shifts, irreversible geopolitical fragmentation, 00:38:26.670 --> 00:38:29.730 augmentation, or uncontrolled technological displacement 00:38:29.730 --> 00:38:32.309 events that have zero historical variance data 00:38:32.309 --> 00:38:35.250 to feed into the traditional MPT model? How should 00:38:35.250 --> 00:38:37.590 we redefine our concept of beta to account for 00:38:37.590 --> 00:38:39.309 the systematic risks that are currently invisible 00:38:39.309 --> 00:38:42.230 to the mathematical models of 1952? Think about 00:38:42.230 --> 00:38:43.849 that as you consider how much weight you put 00:38:43.849 --> 00:38:45.469 on the past when predicting the future.