ORIA (Orthogonal Risk Integrated Alpha) Explanation Pt. 1 - Joshua Kyan Aalampour — Transcript

Full transcript

1. 0:00Hello everybody. My name is Joshua
2. 0:01Kananpour. This is more of an
3. 0:03experimental video. I want to talk about
4. 0:05something that I've been working on for
5. 0:06the last year, which is something called
6. 0:08Oria. Oria stands for orthogonal wrist
7. 0:10integrated alpha, and it is a
8. 0:11quantitative trading architecture. So,
9. 0:13the idea is not just to create one
10. 0:15trading strategy. The idea was to create
11. 0:17an ensemble of trading strategies that
12. 0:19can interact with one another one
13. 0:20another intelligently. So, the basic
14. 0:22architecture is something like this. We
15. 0:24have a couple of different layers. We
16. 0:25have the [music] strategy layer, and
17. 0:27then we have orthogonalization, and then
18. 0:29we have an allocator, and then we
19. 0:31finally have a risk layer, and then an
20. 0:33execution layer. The reason why I wanted
21. 0:36multiple layers here is because it can
22. 0:38be too risky to just solely rely on a
23. 0:40single decent signal, right? Like in
24. 0:43theory, you could have a model that
25. 0:44looks really nice in isolation and still
26. 0:46end up with something that's maybe, you
27. 0:47know, too correlated, too concentrated,
28. 0:50or, you know, just like structurally
29. 0:51fragile. So, how this works more
30. 0:53technically is that each sleeve produces
31. 0:55a raw desired book. And right here I
32. 0:58will write the three, for example,
33. 1:00right? And with those sleeves, we stack
34. 1:02them together. So, this is our stacked
35. 1:05sleeve object here. Okay, so let's think
36. 1:07for a second. You might ask, okay, why
37. 1:09not just combine them directly? The
38. 1:11answer is pretty straightforward. It's
39. 1:12just pure risk management. So, you know,
40. 1:14two sleeves can easily be expressing the
41. 1:16same underlying risk, but maybe in
42. 1:18different [music] language. So, before
43. 1:19we even think about capital allocation,
44. 1:21the purpose of the stack is to strip out
45. 1:23common factor exposure and remove that
46. 1:25cross-sleeve overlap. So, this is a good
47. 1:27thing. This is kind of loosely taking
48. 1:29the saying, "Don't put all of your eggs
49. 1:31in one basket," and just formalizing it
50. 1:33mathematically. Uh we want to
51. 1:34deduplicate risk before allocation. So,
52. 1:37what we're actually doing is taking each
53. 1:38raw strategy and projecting it into a
54. 1:41factor neutral version, and then
55. 1:42decorrelating it in a risk metric. So,
56. 1:45this right here is our raw desired book.
57. 1:47Uh this right here is our risk met-
58. 1:50>> [music]
59. 1:50>> risk I can't speak, oh my god, is our
60. 1:52risk metric that we're going to use to
61. 1:54measure distance, and then um this
62. 1:56condition here is just to enforce common
63. 1:59risk neutrality before we actually
64. 2:00combine the strategies. So, this is the
65. 2:02orthogonalization step. Uh we're not yet
66. 2:04deciding how much money each strategy
67. 2:06should get, you know, instead we're just
68. 2:08making sure that we're not double
69. 2:09counting the same risk before we
70. 2:10actually feed these into our allocator.
71. 2:12So, I'm going to end part one here, and
72. 2:15uh you can continue in part two because
73. 2:18I don't want to make this video too long
74. 2:19for social media.