kbennatti

Love this!

My personal experience supports a lot of this. I’ve become more practical and less idealistic with age and that’s mostly been a great thing for me.

See response above. And not just fraud detection. We plan to expand to other areas including credit/bankruptcy, ESG and tracking macro changes and trends

A few reasons but the primary one is impact. Our mission is corporate accountability through information transparency. Once you become a hedge fund your incentives for information dissemination become warped.

We don't sell to filers. Ever. Got to protect against conflicts of interest. That's one of the reasons we don't have a true free version of the product.

That said, we're using language models so replacing words with synonyms won't evade the model as long as its expressing the same thing.

We love footnoted! Michelle Leder of footnoted has been really supportive of our work. She recognizes that it's an important area for tech innovation. footnoted isn't operating right now but when it comes back, we'll support her and collaborate if we can.

We're often compared to AlphaSense, Sentieo and InsiderScore but our product is pretty different. Competitor products focus on sentiment and linguistic metrics or search etc, not on extracting and organizing important textual content.

Yeah, people have been doing sentiment analysis of press releases etc for years now and very few corporates bother trying to use software to test/counteract it

You've hit the nail on the head (mostly)...but I'm going to say no more because its part of our secret sauce

I love that. I agree until the "nothing" bit. Some things do get priced in. You can see specific stocks "react" to news, adjusted for market returns etc. ESG factors do appear to be getting "priced in" as well

Seconding Suhas here. Super rad

So true

It's disclosed in their filings e.g. "Among the vendors were a director of the Company and an entity controlled by such director"

At lot of egregious things are disclosed on page 101 of a filing but they get missed because these filings are so long and deathly boring

Thanks for website edit! Fixed.

We don't rely on XBRL for parsing. It's not very consistent/reliable and its mostly for numeric content. We've definitely considered integrating ratios both into our dashboard. It isn't a current priority because ratios are already well supported elsewhere.

Aw cool! Can I hang out with your Dad? ;) We do pick up on overly promotional/jargon-y language to some extent. For the most part, however, word lists haven't worked for us.

just a sample

in-kind services is always a red flag because it's easy to fudge the accounting

Nikola red flags for you! (all algorithmic)

1. "For example, in September 2020, our founder and former executive chairman, Trevor R. Milton, stepped down from his positions with us."

2. "During the fourth quarter of 2020, the Company ceased operations related to the Powersports business unit in order to focus on the Company's primary mission of commercial production of semi-trucks and construction of hydrogen fueling stations."

3. "As of December 31, 2020, we have $46.3 million of prepaid in-kind advisory services remaining which is expected to be consumed in 2021 and will be recorded as research and development expense until we reach commercial production."

Yes, it did. BUT here's the thing -> Nikola was first a SPAC and by the time it filed its first filing as Nikola, the Hidenburg report was already out so its not a true example of our algorithms preceding short reports. Our algos did beat Hidenburg to the punch on a number of other companies though.

A bit outdated but check this out - https://bedrock.substack.com/p/bedrock-ai-vs-activist-shorts

Ooh I love MTL. That's the first place I lived in Canada. Great q. We used SEC enforcement actions related to fraud as our gold label (fairly common practice in academia). Really key thing here is that you need to be careful about what years you're using for training because if you include years that are too late in the fraud cycle, you end up with significant target leakage e.g. the filings we'll say, "we're being investigated for fraud". We ended up manually reviewing all of our data/labels. It took over a month. We also use settled class action lawsuits as silver labels. Plus a few other more frequent labels as bronze labels

Interesting. Potentially? How do we get in touch?

Our product has two very distinct outputs 1) red flags and 2) a risk score. Our primary value-add is the red flags which are a qualitative input into an analysts process. We find the information you need to see without having to spend 3 hours reading nonsense. The two components get tested very differently. The red flags explain (sort of) the risk scores but not in a truly explainable fashion. The risk scores are optimized for precision and approximately 1 in 3 companies with a "fraud" score will end up with an SEC investigation or equivalent.

Yes! the NLP research we've done can be adapted to other tasks.

We don't sell to filers. Ever. Got to protect against conflicts of interest. That's one of the reasons we don't have a true free version of the product.

That said, we're using language models so replacing words with synonyms won't evade the model as long as its expressing the same thing.

Or this? John Rohner, claimed that he'd developed tested and patented a "plasma engine" fueled by inexpensive and abundant noble gasses. He also claiming that he graduated from Harvard at 14 and had 3 Phds from M.I.T. Somehow he managed to dupe 98+ investors

Like this? This Oxford City Football team, a boiler room operation, duped people by saying that they were bound by a voice recognition and proved it by using the phone to make some beeping sounds. https://www.sec.gov/litigation/litreleases/2017/lr23869.htm

Re: alpha -> our focus is on extracting red flags that are similar to what a forensic accountant/analyst would find. AI-assisted research rather than AI driven. trading on fraud signals alone is pretty hard, you need another event. we havent done quant testing of our risk scores in a while. We definitely should do proper quant backtesting though.

RE: backtesting we use SEC enforcement actions related to fraud (10b-5) as our gold label. That said, there is no gold label for absence of fraud so our backtested metrics are probably slightly better than they seem.

We've never tried to score a short fund on their precision/recall but unofficially, Hindenburg Research has the highest concordance with our models in deployment.

> Humans too, you're describing at least one comment in every HN thread about an SEC filing :D

LOL. Yup!

Great q. Sino-Forest is an out-of-sample test so our models didn't technically "learn" from it. That said, very valid comment. Historical testing only goes so far. Assessing whether our algorithms work in deployment has been cool. Check out some of our live, in deployment examples here - https://bedrock.substack.com/p/bedrock-ai-vs-activist-shorts

Sweet! The world needs a Bloomberg alternative so props. We don't have an API atm. We're focused on supporting human analysts. The meat of the product is the red flags which are textual and not quantitative.