market folly

Eric graduated from Northwestern with a PhD in economics and wrote his dissertation on cross sectional stock returns and volatility. Prior to joining Telluride as a hedge fund manager in 2004, he had been using strategies that drew upon his education, previous work running his own fund and a fund, Deephaven.

Eric resigned his position at Telluride in September, 2006. Several months later, Telluride initiated a lawsuit claiming that strategies used by Eric belonged to Telluride. The claims in the lawsuit would require a court to determine the nature of the components of the strategy Eric had been using and decide who owned them. This is more complicated than it may appear.

Consider this analogy. Baker E goes to work baking sugar cookies for Bakery B. E, who’s been a baker for ten years, has a favorite recipe calling for flour, butter, sugar, eggs and baking soda. During the next few years, E tinkers with the ratios of ingredients and experiments with chocolate frosting and colorful sprinkles, but never deviates significantly from the basic recipe.

Then one day, Baker E decides to leave Bakery B and open a Cookie Shoppe C in another town. Bakery B initiates legal action to prevent E from operating C, arguing that E’s cookie recipe will inevitably be derived from privileged information gained while working for B.

In response, Baker E argues that his recipe is a standard sugar cookie recipe, using common ingredients. He argues that B cannot own the sugar cookie constituents (sugar, flour, butter, etc.), and that B needs to define the specific recipe in its complaint. Bakery B argues they will provide that, after full discovery has been completed (which could be a few years).

In a hedge fund “trade secret” case, the “ingredients” are variables used to construct a fund manager’s strategy, and the use of these variables may vary. In contrast to cookie baking, the manner in which the variables may be used are not obvious. Because the analysis is not intuitively understood by lay persons, any assertion by the complaining party may appear tenable.

Broad financial concepts of profitability, volatility, and mean-variance optimization and virtually any financial ratio or indicator may be particularly troublesome. While lawyers and courts are generally familiar with cookies, they are not familiar with hedge fund strategies. They must differentiate whether certain variables and their use are in the public domain or within the definition of a trade secret, and what kinds of ‘ideas’ fall under the confidential agreement. They might also need to decide whether the variables were used by the ex-employee prior to his work with the complaining party.

In Telluride Asset Management LLC v. Eric Falkenstein, Telluride claimed that Eric was violating his confidentiality agreement, which included trade secrets, but also ‘all inventions, discoveries, computer software programs, trade concepts, designs, patents, ideas … conceived or developed by Employee’ during his employment. Eric claimed his only planned overlap with his former work pertained to common tools, ones he used for a decade prior to his work at Telluride.

Eric’s story received some media attention from Megan Barnett in May of 2008. She wrote in “Cudgel Over the Quants,” Portfolio.com:

“Eric Falkenstein isn’t your typical 42-year-old hedge fund manager. Instead of trading stocks all day or courting new investors, he spends his time updating his blog, researching equity strategies, and talking to his lawyer. He’s a hedge fund portfolio manager who is legally restrained from managing hedge fund portfolios.

But Falkenstein didn’t embezzle funds, swindle unsuspecting investors, or violate insider-trading laws. Rather, he quit his job one September day in 2006 and he hasn’t been able to work since…

Welcome to the murky world of hedge fund trade secrets, where your likelihood of getting a new job may be directly related to your employer’s inclination toward litigation. These types of trade-secret suits are generating a controversy in the hedge fund industry. Is the litigation little more than a bullying tactic to keep valuable employees from heading to a competitor, as the blogger Equity Private suggests? Is the specific knowledge of trading strategies one gains at a hedge fund legitimately unusable in any future endeavor? Or are traders stealing secrets with the hopes of making more money from them someplace else?”

Trade secret suits are especially difficult to defend in many states because the trade secret does not have to be defined and can be changed after discovery ensues.

As anonymous blogger Private Equity commented on Eric’s case in “IP Litigation Arbitrage Tactics,” April, 2008:

Work for a hedge fund, perhaps as a quant, depart and try to work in the field again. Instead of attempting to directly enforce a non-compete agreement, the hedge fund might bring an intellectual property case based on trade secrets. Now the non-compete provisions, which on their own are not likely to work well, merely become further evidence of the former employee’s bad faith.

If you are a clever hedge fund, you will then seal your complaint, after all, it contains sensitive trade secrets. What sort of secrets? Let’s take an example, perhaps from a temporary restraining order [TRO] granted to Telluride Asset Management against their former employee, Eric Falkenstein….

And how might you, as a hedge fund in this position, respond to the suggestion that these factors and mean-variance optimization might be rather obviously in the "public domain" and therefore beyond enforcement as a "trade secret"? First, in your complaint, indicate that your trade secret is "a specific application of mean-variance optimization that will be defined after discovery." That will delay the argument until a fishing expedition can be conducted and a connection made. Then, insist that your former employee disclose the entire model to prove it doesn’t infringe…

Here’s the great part: As a hedge fund, you don’t even need to be granted the TRO. You merely file it and get discovery going. The effect is the same…

Discovery is key in the tactic. The hedge fund can insist on a core dump of, for example, every email the former employee has written since employment, including material from personal email accounts.

Well, sure, that’s painful, but no big deal. The former employee can just go back to work and make money to pay the defense lawyers. Sure, if you can find an employer willing to retain you and your methods when it might mean unlimited liability for every dollar you make for them thereafter…

For Eric, the experience was emotionally exhausting and extremely expensive. Telluride’s lawyers examined 10 years worth of data from his hard drives, including personal computers, emails, every aspect of Eric’s life as recorded on his computer and in trails of online activity. Because the specific ‘ideas/designs/trade secrets’ were not defined at the outset, lawyers searching for evidence of stolen IP could identify it after seeing all his data. Again, the search, productive or not, could take years.

"The nasty thing about intellectual property cases is that one can use it to start discovery on a broad scope of information, and then generate a post hoc definition of what is covered. The key is that overbreadth in technical matters is not obvious. For example, one can say, "he took the secret of mean-variance optimization", and by the time the court figures out such a claim is absurdly overbroad, a more tenable claim can be made, such as ‘a specific application of mean-variance optimization that will be defined after discovery’. To the defendant, both claims place all his activities as potentially poisoned, so the effect on the defendant’s ability to work is unabated. But to the court, the latter works as long as he has some deleted files on his home computer related to his work, and ‘related’ can be rather boring stuff that the court does not recognize as well-known, such as a spreadsheet with S&P returns and Excel formulae." Eric, Goldman Quant Case Continues, Falkenblog, August, 2009.

In addition, Eric was unemployable as a hedge fund manager while living and paying for his legal fees out of savings. These factors ultimately led to Eric filing a counter-claim and the suit being settled. Eric notes, "The settlement gives me complete freedom, but there’s a stain there. I could have burnt more money litigating, but my chances of receiving damages that would cover legal expenses and opportunity costs were small, and a final judgment would have given me no greater ability to ply my wares than the settlement.”

Eric: At our final hearing where I got three counterclaims inserted into the case, the judge strongly suggested mediation. So, with the judge’s encouragement, we entered mediation and settled our dispute in a few hours. We agreed not to re-file our claims, and that I can use anything I used at Telluride going forward.

Eric: The main one was tortuous interference, i.e. that Telluride asserted its IP and contractual rights in bad faith and in a knowingly overbroad manner that directly affected a specific business deal.

Eric: In response to an inquiry by Telluride as to my proposed venture, I mentioned that the only similarities were common tools that I thought I had full right to use, factors like profitability, changes in equity, volatility, accruals, and the process of mean-variance optimization. I thought being explicit would make it clear I merely wanted to do what I had done previously, using factors I used before, factors that are well-known in the academic literature. This was incredibly naïve on my part. My disclosure was used as a prime exhibit in their complaint against me as ‘proof’ I was violating the confidentiality agreement. That is, I admitted using factors used at Telluride, and supposedly after they showed me how these factors work, my future use could not help but be derived from this privileged knowledge. To someone who has read the literature on these common factors, and how they were applied, this seemed insane, but to a judge it was not obvious, and once IP litigation starts, as a defendant you’re in the penalty box.

Eric: There’s a sealed list of concepts they assert are their confidential information in our settlement. If I work with you I can explain in detail the specific concepts in dispute, as defined by Telluride, but not otherwise. It’s been several years, and I’m always working on new ideas, making these specifics totally uninteresting to me, so it’s really a moot point. The key is, I’m no longer a liability, because I bought a ‘perpetual license’ to these concepts (I paid secret amount ‘x’ for this license). I have a right to use anything Telluride could claim via my license.

Eric: I cannot discuss my track record at Telluride. Thus, my largest datapoint as a portfolio manager was a black hole. Furthermore, larger institutions are especially wary of managers with ‘hair’ on them via working at large firms, or having engaged in litigation.

Eric: Well, you can insulate yourself in various ways—negotiating ex ante that you can use your track record going forward, actually throw away any computers used contemporaneous to your employment, examine a potential employer for previous litigation tactics–but ultimately, it’s about judging people’s reasonableness. Like judging a date on how they treat the waiter, be aware of unreasonable behavior, and then ratchet your precautions accordingly. You don’t want to start every relationship assuming the worst but you need to be careful. The key to having good relationships is picking reasonable partners, as opposed to constructing a bunch of formal legal agreements. Good faith goes a long way.

Eric: First, I would make firms define their trade secrets or specific concepts subject to any confidentiality agreement prior to discovery. I would also like a ‘loser pays’ rule, because you can win your lawsuit and still lose, because you cannot count on getting damages from a counter-claim. This happening is probably unlikely.

But here is a novel item worth noting. You can see the model I proposed when I came to Telluride, it is in the Hennepin County documents (unsealed). It presents a rather straightforward model based on several exclusionary rules, or sequence of sorts. For example, to simplify, say I arrived at Telluride with a strategy in which I construct a long portfolio by first targeting only firms with market caps greater than $500MM, then take those top 200 companies with the highest cash-flow, and then take the top 100 within that which had the best momentum. Now, Telluride basically argued I could not use these factors because they were inevitably derived from special knowledge acquired while at Telluride. The court eventually ruled they could not own these factors (profitability, accruals, capital issuance, and volatility) or processes (mean variance optimization) in general. So Telluride then merely said, we own them in a particular usage to be defined after discovery, and the court let them proceed on that path.

Implicitly, the court anticipated some set of applications of this logic would be forbidden by me, but it was never clear to me what they could then own in practice. That is, if I took the model above, and say changed it while I worked at Telluride so that I only used companies with market caps greater than $600MM, and the top 220 highest profitability, and then top 100 by momentum, how does that modification affect what I cannot do outside the firm given our confidentiality agreement? What about using $700MM as a cutoff, and 300 top cash flow firms? What if I used a new algorithm that, say, transformed cash flow and momentum into percentiles, then added those numbers together, and chose the top 100 for my longs? How is the line drawn in these cases for parameters and other algorithms use these same factors? It was never clear to me what the end game would look like, because if I enter a firm with an algorithm which uses a set of inputs, I clearly can use that, but then I make modest changes. If they own a family of things extrapolated from those changes, how is this determined? In an algorithm, you often can’t simply ‘split the difference’.

Or to take another example, I planned on using mean variance optimization going forward, and Telluride objected. After some months, the court agreed with me that they did not own this concept in general, but they could own it in some way they could define after looking through all my hard drives. Mean-variance optimization is very well known, and basically a way to generate portfolio weights. I could not imagine the provenance of the parameters could affect whether I use this very well known technique. It’s a well defined problem, so the solution is trivial. You can even buy software that does it.

I wonder how often people are constrained in this way, prevented from using common tools because these were used while working at a firm where they had signed a confidentiality agreement. You don’t hear about it much, but litigation databases are hard to search, and people don’t like to talk about it.

Eric: Finding Alpha is about the search for ‘risk adjusted outperformance’, or alpha. My main argument is that the ‘risk adjustment’ is trivial because risk and return are not correlated, so the expected return on most assets should be the same. The absence of a risk premium in so many domains is not an anomaly, but an empirical fact, and I present a novel scope of information relevant to this. Further, this pattern is a consequence of a modification about how people internalize their wealth, which is relative to others, as opposed to comparing to having absolutely nothing. So it’s mainly an argument against the conventional theory that risk generates a positive return premium, and it gets into technical issues about utility functions. The practical application is you should only expect to make a return above, say, the BBB libor rate, by being smart, not merely taking some measure of risk.

This is a pretty profound difference from the standard theory. Some rather straightforward investment strategies are implied as having higher returns for the same level of volatility or beta if this is true. Also, as alpha is a risk adjusted return, and ‘risk’ is not unambiguously defined, there’s a lot of room for shenanigans.

Eric: The fact this is considered a criminal, not a civil matter, highlights the political muscle of Goldman. From what I have read this ex-Goldman employee seems guilty, in that he came to Goldman with no trading experience, left with lots of code explicitly mentioned in various agreements, and suddenly was worth $1.2MM to a new employer. It seems rather incredible to believe be this programmer developed $1.2MM worth of alpha without appropriating Goldman’s IP.

Ilene: Did the financial meltdown of last year surprise you? Did your models (if you were still applying them) predict any kind of sudden market decline?

Eric: I did not see it coming, but the collapse tended to hurt firms that statistically underperform over the long run, and in that way is consistent with my models. However, the rebound this year did just the opposite, where the longer run losers have done especially well this year. That’s not inconsistent with my models, but in bear markets, the bad stocks (high volatility, low profitability, high capital issuance, negative momentum) do really, really badly, and the good stocks relatively better. In the snap back, however, the performance is strongly anomalous, though temporarily.

Eric: Up for stocks, down for Treasuries, up for corporate debt. I don’t like a lot of longer term signals, as I think government is ascendant in all sorts of bad ways that will hurt productivity growth in the long run. But that’s a long run effect. In the short run, people were expecting another Great Depression and that’s not going to happen, so banks and REITs have a lot of room for recovery.

Eric: I just took a job with a local trading company, working on various quantitative projects. It’s a local option market making firm.

Eric: Sure, new opportunities are always exciting. When I stop getting excited, I’ll retire, or become a risk manager :-)

****