Can AI Understand a Distressed Credit? - Harneet Kaur, Axim Intelligence

AI Engineer · 14 min · 148 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

1. 00:13Thank you to everyone who came.
2. 00:14I am very pleased to deliver this report.
3. 00:18So, a brief introduction about me.
4. 00:22My name is Harnit.
5. 00:23I am the founder and CEO of Axon Intelligence.
6. 00:28At Axon, we create an intelligent layer for investing in distressed loans.
7. 00:33This is also what we will talk about today.
8. 00:37The title of my talk: Can AI understand problem loans?
9. 00:45I think the first question that arises for everyone who is not familiar with loans is: what does a problem loan actually mean?
10. 00:55Credit is pretty simple.
11. 00:57It is simply a bond or loan from a corporation.
12. 01:01A Coca-Cola bond is a loan.
13. 01:06And what does it mean to be under pressure or in a difficult situation?
14. 01:10Is n't that right?
15. 01:14Pressure arises if a company has some underlying risk that could cause them to run out of money or be unable to repay their debt.
16. 01:25Every year, a certain portion of the credit market faces some level of pressure or problems.
17. 01:34This may be caused by cyclical problems.
18. 01:38Let's say the construction market is in decline, the housing market is in decline.
19. 01:42This may be caused by operational issues.
20. 01:45For example, if you are a restaurant business and there is a large-scale product recall.
21. 01:49This may also be caused by long-term structural changes .
22. 01:53If you are in software today and are facing the impact of AI.
23. 01:58Some companies overcome difficulties and recover.
24. 02:01Some companies face challenges and find themselves in difficult situations.
25. 02:06What does a difficult situation mean?
26. 02:08In reality, it's not that simple.
27. 02:10Bankruptcy is one of the consequences of a difficult situation.
28. 02:17Liability management measures are also possible—I can explain these terms in detail, but just know that this is another option, as well as an out-of-court restructuring or an agreed exchange.
29. 02:30So, next slide .
30. 02:32As you can see, it's not that simple, is it?
31. 02:39Historically, the issue of predicting problems has been quite binary.
32. 02:43Bankruptcy could be predicted : either you predict default or you don't.
33. 02:51Today it has become much more difficult.
34. 02:57Many traditional machine learning methods simply can't keep up with this, nor can the assessments that were important 10, 15, or 20 years ago.
35. 03:10Um, uh, the main problem is that Altman's Z-score has become very static in a situation that changes, you know, sometimes monthly, quarterly, or annually, and in a much more opaque
36. 03:22market.
37. 03:24So, if you've been following the credit market, the growth of private lending and some other sectors has led to, you know, a lack of pricing.
38. 03:33On the other hand, there is a very high-quality method of understanding and analyzing problem loans, when you, as an investor in problem debt, can analyze financial statements, understand what is happening on the part of the lender, and read loan documentation.
39. 03:49Basically, you know, all the things a typical hedge fund or private equity investor would do to understand credit.
40. 03:57This process is extremely manual and, of course, very difficult to scale.
41. 04:01Isn't that right ?
42. 04:03Now the question is , you know, part of the founding of Axon and the question that I've been thinking about is how far AI can go in understanding distressed
43. 04:16credit and what paths can be taken.
44. 04:19And is it possible to systematize what is actually a rather complex process, right?
45. 04:27So, what does it mean to understand a non-performing loan, right?
46. 04:34As investors, we have many questions that we are trying to find answers to.
47. 04:41Things you think about from a financial perspective, right?
48. 04:47Do you know what liquidity is?
49. 04:49What is the capital structure?
50. 04:51And then there are legal questions.
51. 04:54How do you know what the structural priority of the debt is ?
52. 04:57Isn't that right?
53. 04:58That is, do you have a higher or lower priority?
54. 05:02And then there are operational questions, like: how do you understand business?
55. 05:05What are the revenues there?
56. 05:07Will revenues go up or down tomorrow, based on fundamental, you know, fundamental business analysis.
57. 05:14Is n't that right?
58. 05:16Right now, what we're seeing in the market, if you follow Pro Go, Harvey, or some of these big AI companies, they're actually thinking about how to compare all these different
59. 05:27issues, right?
60. 05:29For example, how well does AI answer the question of which portion of debt is structurally senior?
61. 05:37And then do thorough comparative testing and make it, let's call it, a big financial benchmark.
62. 05:45But I wanted to think about it a little differently, didn't I ?
63. 05:50Because I was thinking this way: the real understanding of credit, especially distressed credit, is whether you can predict what will happen next?
64. 06:00Isn't that right ?
65. 06:02Can I figure out and get a probability distribution of which way this will go in the future?
66. 06:08Um, so I thought this was a pretty interesting exercise.
67. 06:15The first step, you know, what you should do is take all the, you know, I chose the company.
68. 06:20So, I chose iHeartMedia, a company that filed for bankruptcy in 2018 .
69. 06:27So she, you know, she went through all these bureaucratic hoops.
70. 06:31Um, she also had another stressful transaction in 2024.
71. 06:36So this is a real-time situation.
72. 06:39This is happening today.
73. 06:41Nobody has a real, you know, answer to what the future holds, right?
74. 06:46Um, there is no absolute truth.
75. 06:49But what I did was this: let's throw this into the smartest, you know, models.
76. 06:53I used Astra 6 on Max Intelligence and Opus 5.5. And the conclusion that I found very interesting is that they disagreed.
77. 07:06They didn't agree on what the probabilities of the future would be, given the structure or context of the data, right?
78. 07:17And so, if you look here at the agreed restructuring, Astra thought it was the best option for bankruptcy, and Opus thought it was the best option, right?
79. 07:28Now, how do you tell who is right and who is wrong?
80. 07:34You can't, because I don't know what the probabilities are, right?
81. 07:38I don't know what will happen.
82. 07:40I don't know if it will be a bankruptcy.
83. 07:42I don't know if this will be an out-of-court restructuring.
84. 07:45I do n't know if this will be a refinancing.
85. 07:48I have my own judgment on these things.
86. 07:50I would say I'm leaning towards Astra 6, but, you know, we're all human.
87. 07:56And that's a pretty interesting conclusion, right?
88. 07:58Like, okay, wait, they disagree.
89. 08:03Um, so the second thing I thought about was: okay, how can I at least improve their process assessment, right?
90. 08:11So I can see whether I agree or disagree with different assumptions.
91. 08:18Um, and so, essentially, you've created a kind of benchmark where we test every level.
92. 08:29Obviously, mistakes can be made at every level of the LLM, right?
93. 08:33Regarding the level of data extraction, I would say that with public companies this issue is already pretty much resolved.
94. 08:39I would say with private loans, and also when you have Excel and PDF documents, it can still be a challenge, right?
95. 08:47Then you have the reconciliation or data output level.
96. 08:50So, extraction is, for example, obtaining the EBITDA indicator.
97. 08:52Uh, the conclusion would be : okay, now you have EBITDA and you have debt, so you have a credit load.
98. 08:58And then you have judgment.
99. 09:00Okay, now you have a credit load.
100. 09:02Is this a problem loan or not?
101. 09:07What is your Oh.
102. 09:19Uh, is there any...okay, I'll just, I'll just keep talking.
103. 09:27So, uh, actually, the question is , is it like my computer has shut down?
104. 09:43Uh, excuse me, just one second .
105. 09:47I'll tell you I have a charger, but it's not here.
106. 09:57Uh, yes.
107. 10:05Okay, I'll just, I'll just talk.
108. 10:08This is normal.
109. 10:13Oh, okay.
110. 10:14It works.
111. 10:19OK.
112. 10:21So, therefore, you have a level of judgment , right?
113. 10:25And then, uh, then, then you can start, okay , if you can evaluate each level, right?
114. 10:32When you get to the prediction level, you've made sure the data points are right, you've made sure the, uh, level of agreement is right, you've made sure there's some kind of reasonable judgment,
115. 10:45right?
116. 10:46And then , once you're confident in those things, then you can move to the forecasting level, which is really how, you know, you as a human analyst think about all of these steps, right?
117. 10:57You, you, if you base a judgment on a bad data point, it's not going to be a good judgment, and that's, that's just how it works, right?
118. 11:07So, uh, Op.
119. 11:14Here we are.
120. 11:17So, overall, what we found, and maybe this isn't surprising, but, uh, I think it was interesting, is that once we created a structured reasoning process, you , you actually got both intellects,
121. 11:29both models, to actually converge on what they thought the future probabilities for the business would be, right?
122. 11:42So now, instead of talking about a 62% bankruptcy with Opus and a 25% with Astra, now both of these models say that there is actually only a quarter, or a 25% chance of bankruptcy, and both models agreed that the path of concerted
123. 11:56restructuring is actually the more likely path for the business.
124. 12:02So, what are the main conclusions here, right ?
125. 12:07Uh, one, I would say, first conclusion: if you give questions that cannot be verified , right?
126. 12:14With the same structured context, organized context, as humans, I mean , do these models have an advantage in prediction?
127. 12:25And you have to say "no", right?
128. 12:28Because they are, in fact, even agents— they are people, right?
129. 12:33In some sense of the word, and I won't get into an argument about that .
130. 12:37Uh, the second thing is, but a structured reasoning process can force two models or, you know, different intellects to actually come to the same vision of the future and a similar understanding of the business and the path it
131. 12:55can take.
132. 12:58Hmm, but is this the right way ?
133. 13:02We don't know, do we ?
134. 13:04Um, and is that really the single most accurate measurement, right?
135. 13:07Otherwise, any historical analysis would create a certain distortion by predicting the future.
136. 13:13Um, but I think the really interesting thing that I was thinking about is that convergence is also quite interesting, right ?
137. 13:22Because if you can systematically build a process that actually converges to the same probabilistic outcomes.
138. 13:30This, in fact, is the market price.
139. 13:33Um, and that's why I thought that was an interesting conclusion.
140. 13:38Um, so this is the last one.
141. 13:42And now briefly about me.
142. 13:43So, my name is , as I said, Harini.
143. 13:46Um, building an Axon.
144. 13:48Um, a little bit about my background.
145. 13:49I spent 10 years in finance, um , the last few years as a distressed and special situations investor at a, uh, private equity/ distressed asset hedge fund.
146. 14:00Um, here's the QR code if you're interested in Axon.
147. 14:04Um, and that's all I had.
148. 14:06Thank you.