or Connect
Styleforum › Forums › Culture › Business, Careers & Education › Starting an investment business
New Posts  All Forums:Forum Nav:

Starting an investment business - Page 16

post #226 of 273
Quote:
Originally Posted by NameBack View Post

Oh and as of end of April we're up 12%, since inception, with the average hedge fund up 4.5% over the same period.

 

of the same strategy/category or in general?

post #227 of 273
Congrats, will be monitoring this thread periodically.

Also PMed you for the pitch book wink.gif
post #228 of 273
Thread Starter 
Quote:
Originally Posted by chogall View Post

of the same strategy/category or in general?

In general.

Haven't been able to find a breakout of just quant funds.
post #229 of 273
Thread Starter 
Going to the SALT conference in Vegas on Tuesday of next week. One of the two biggest events in the hedge fund world each year, tons of investors. Two events there that I know I can get into and the rest I'm going to have to crash. Should be fun!
post #230 of 273
Some advice from a former HF analyst / current bank trader.

- always report returns at 1x leverage. Most investors will want to set their own leverage / vol target
- if someone distributes this for you, that likely makes them a fiduciary which means then can (and will) get sued if things go wrong (and they will if you promise anything like 300%)
- real time / out of sample tests are useless. Trade small amounts of real money, don't bother paper trading

I don't believe this model is nearly as useful as you think but you seem intent on disregarding any comments along those lines so I won't take the time to explain why.
post #231 of 273
Thread Starter 
Quote:
Originally Posted by Cantabrigian View Post

Some advice from a former HF analyst / current bank trader.
- always report returns at 1x leverage. Most investors will want to set their own leverage / vol target
- if someone distributes this for you, that likely makes them a fiduciary which means then can (and will) get sued if things go wrong (and they will if you promise anything like 300%)
- real time / out of sample tests are useless. Trade small amounts of real money, don't bother paper trading
I don't believe this model is nearly as useful as you think but you seem intent on disregarding any comments along those lines so I won't take the time to explain why.

I'm not intent on disregarding -- I think I've answered most questions on the nature of the model pretty even-handedly up to this point. Most of the concerns people have voiced haven't really applied to our methodology, and IIRC most of the conversations on the issue have ended with the concerned party's fears being assuaged.

I'm curious what your comments are given your background in the field.

edit: Also I'm curious as to why you would say that out of sample tests are useless. As long as the methodology is rigorous, how is it different substantially from a live test? Although, you'll note we have been trading small amounts of real money for almost six months now in order to build a track record.
post #232 of 273
Quote:
Originally Posted by NameBack View Post

I'm not intent on disregarding -- I think I've answered most questions on the nature of the model pretty even-handedly up to this point. Most of the concerns people have voiced haven't really applied to our methodology, and IIRC most of the conversations on the issue have ended with the concerned party's fears being assuaged.
I'm curious what your comments are given your background in the field.
edit: Also I'm curious as to why you would say that out of sample tests are useless. As long as the methodology is rigorous, how is it different substantially from a live test? Although, you'll note we have been trading small amounts of real money for almost six months now in order to build a track record.

I'd imagine it's because unless you can somehow account for how your own actions affect the pricing of the securities you trade, paper trading will always result in an upwards biased result unless you're talking about trading blocks in such small amounts that other players can't tell who it is and whether or not they can pip you.
post #233 of 273
Quote:
Originally Posted by Nereis View Post

I'd imagine it's because unless you can somehow account for how your own actions affect the pricing of the securities you trade, paper trading will always result in an upwards biased result unless you're talking about trading blocks in such small amounts that other players can't tell who it is and whether or not they can pip you.

Unless you get a Buffett pass with the SEC to delay reporting...worship2.gif
post #234 of 273
Yea - I didn't write that third point correctly at all. Paper trades are obviously useful to you for testing / further R&D but investors won't put much (any) stock in that. If you're looking to move beyond friends and family money, just to give you an idea of how hard it is to get institutional money, you've had hardly any big bank prop traders spin off successfully in the past four years with the exception of Phibro / Andrew Hall. Because investors discount unaudited results completely.

There are several reasons - from the naturally conservative nature of those investors to the fact that you're very levered and simulators don't do a good job of accounting for cash management which post Bear Stearns is absolutely essential. Nobody gets term funding now or if they do, it's at a price.


On the model, there are a few obvious red flags - like obvious and complete deal killers for institutional allocators / investors
- If you're up 12pct MTD, you're levered and the some. I'm assuming at least 4x and finger in the air, I'd say more like 10x. You can barely get away with that for some relative value / fixed income strategy but that's too much on equities.
- Related to that, look at what you make 1x or better .75x levered. Anyone can borrow money, what people care about is beating the relevant benchmark on an unlevered net return basis and/or on a vol-adjusted basis.
- Assume no one will tolerate more than a 10% capital loss rolling YTD - i.e. being down no more than 10% YTD in any calendar year. What do returns look like with a stop like that?
- You performed best in 2000 and 2008. Two exceptional years for the market. That immediately looks like overfitting.
- You need some story about what your edge is. A lot of people with a lot better resources have failed to match that performance. What are the odds that the proverbial two guys in a garage (even if one knows stats) stumbled upon the winning secret. These are deep markets, how did you find what so many people missed?
post #235 of 273
Quote:
Originally Posted by Cantabrigian View Post

Some advice from a former HF analyst / current bank trader.
- always report returns at 1x leverage. Most investors will want to set their own leverage / vol target
- if someone distributes this for you, that likely makes them a fiduciary which means then can (and will) get sued if things go wrong (and they will if you promise anything like 300%)
- real time / out of sample tests are useless. Trade small amounts of real money, don't bother paper trading
I don't believe this model is nearly as useful as you think but you seem intent on disregarding any comments along those lines so I won't take the time to explain why.

 

+ 1

post #236 of 273
Thread Starter 
Quote:
Originally Posted by Cantabrigian View Post

Yea - I didn't write that third point correctly at all. Paper trades are obviously useful to you for testing / further R&D but investors won't put much (any) stock in that. If you're looking to move beyond friends and family money, just to give you an idea of how hard it is to get institutional money, you've had hardly any big bank prop traders spin off successfully in the past four years with the exception of Phibro / Andrew Hall. Because investors discount unaudited results completely.
There are several reasons - from the naturally conservative nature of those investors to the fact that you're very levered and simulators don't do a good job of accounting for cash management which post Bear Stearns is absolutely essential. Nobody gets term funding now or if they do, it's at a price.

I actually agree with all of this. I don't expect that we'll be able to receive any institutional money for at least our first three years. However, that still leaves open individuals, family offices, FoFs, and Seeders. I think between those options we can scrape together enough to start the business and sustain it long enough to have a true track record that would make institutions comfortable.
Quote:
On the model, there are a few obvious red flags - like obvious and complete deal killers for institutional allocators / investors
- If you're up 12pct MTD, you're levered and the some. I'm assuming at least 4x and finger in the air, I'd say more like 10x. You can barely get away with that for some relative value / fixed income strategy but that's too much on equities.

I think you may have misread my post on our performance. We're up 12% (although currently 15% -- it's been a good week) on YTD, not month-to-date. We are levered, but nowhere near those levels. 1.75x. I see what you're saying with running the numbers unlevered and I think that's actually a good idea. It wouldn't be so terribly hard to incorporate that into our backtesting, so I think that's doable for us. However, with the live fund, we've been running it at 1.75 thus far and I'm hesitant of reducing that halfway through the year. It would make our returns look more inconsistent than they actually are, I feel.
Quote:
- Related to that, look at what you make 1x or better .75x levered. Anyone can borrow money, what people care about is beating the relevant benchmark on an unlevered net return basis and/or on a vol-adjusted basis.
- Assume no one will tolerate more than a 10% capital loss rolling YTD - i.e. being down no more than 10% YTD in any calendar year. What do returns look like with a stop like that?

This is one of our biggest flaws. We have drawdowns larger than 10%. They recover quickly and fully, but they do happen. Some kind of stop would crush us because we need to keep trading aggressively to recover quickly.
Quote:
- You performed best in 2000 and 2008. Two exceptional years for the market. That immediately looks like overfitting.

Actually I said our best years were 2008 and 2009 -- which can easily be explained as a function of the super high volatility of those two years. 09 is better for us than 08. It's not just crash years that we do well in.

Quote:
- You need some story about what your edge is. A lot of people with a lot better resources have failed to match that performance. What are the odds that the proverbial two guys in a garage (even if one knows stats) stumbled upon the winning secret. These are deep markets, how did you find what so many people missed?

First I disagree that there is "the" winning secret. I think in the quant world there are many winning secrets, and many yet to be discovered. But with regards to your question, I think we didn't stumble upon it -- we had the advantage of being from the world of analytics and attacked the problem in that way. We knew predicting price movement directly would be too difficult, and it's too computationally complex, so we decided to track the performance of different-but-related trading strategies instead, and dynamically switch between them. That's the key concept that underpins what we do: it's a much simpler task, computationally, to gauge the current performance of a trading strategy, than it is to predict price movement. I can't speak to the extent that others have or haven't come up with the same concept, but I think we attacked the problem in a very straightforward analytic way -- take the thing you can't do, or that's flat-out-impossible, and find a workaround to reduce the number of variables and the overall complexity.
post #237 of 273
Sorry, I'm confused. Are you trading with real money or not? If not, what is the bare minimum you need in your account to construct the portfolios you've been constructing?
post #238 of 273
Quote:
Originally Posted by NameBack View Post

Actually I said our best years were 2008 and 2009 -- which can easily be explained as a function of the super high volatility of those two years. 09 is better for us than 08. It's not just crash years that we do well in.

The point is the same. Those were two years with one very particular characteristic - very strong trends that (with the exception of Dec 08) that showed very little pullback from the trend. That tends to be the exception and generally backtested systems look overly good because of that sort of thing - it was smalls profitable most of the time and killed it in [fill in very specific circumstance that may or may not ever repeat]. It would be stupid to discount a model because it did well in those years and I'm not saying it's flawed because it did well then. All I'm saying is that it smells like overfitting. Out of sample tests should confirm or rebut that but as I understand it, it's a very trap to fall into.
Quote:
Originally Posted by NameBack View Post

This is one of our biggest flaws. We have drawdowns larger than 10%. They recover quickly and fully, but they do happen. Some kind of stop would crush us because we need to keep trading aggressively to recover quickly.

Bolded part = classic Martingale strategy / fallacy. Whenever you lose, double up and you'll eventually win it back.

Money management is commandments 1 - 12 of trading. You can almost suck at everything else and make money if you're disciplined. Loses like that should scare away investors (who don't share you last name) but it should also worry you - a lot. What's your biggest drawdown? There isn't any reason that you couldn't see two or 1.5 of those back to back. The fact that it didn't happen on the backtest should provide you any comfort in that regard.

I'm interested - sometimes your descriptions make this seems like a trend following model but this makes it seem like a mean reversion one. Or maybe you switch between the two.
Quote:
Originally Posted by NameBack View Post

First I disagree that there is "the" winning secret. I think in the quant world there are many winning secrets, and many yet to be discovered. But with regards to your question, I think we didn't stumble upon it -- we had the advantage of being from the world of analytics and attacked the problem in that way. We knew predicting price movement directly would be too difficult, and it's too computationally complex, so we decided to track the performance of different-but-related trading strategies instead, and dynamically switch between them. That's the key concept that underpins what we do: it's a much simpler task, computationally, to gauge the current performance of a trading strategy, than it is to predict price movement. I can't speak to the extent that others have or haven't come up with the same concept, but I think we attacked the problem in a very straightforward analytic way -- take the thing you can't do, or that's flat-out-impossible, and find a workaround to reduce the number of variables and the overall complexity.

I was being flip - sure there are a million winning secrets. But there are also a million guys with a trillion dollars chasing each and every one of those secrets.

You're trading stocks / ETFs, not options right?

In that case you, what you think you're doing doesn't differ in any meaningful way from predicting price movements (1) because it's just as hard to guess the right strategy / market regime as it is to guess the direction of the market and (2) because the strategies that you switch between only profit off directional bets, you also need to predict direction.

And to the original point, there are tons of people with more resources and more help looking at dynamic allocation / market regime models. You're not trying to do anything new here. You may well have done it better than anyone else but in order to pitch it to other and to not lose your shirt, you should understand what your edge is and be exceedingly careful to stick to that and only that. Super smart people get rolled all the time. The other side of that coin is that you still have some guys who barely went to college who are successful traders - they know what exactly it is that they know and stick to their knitting.
post #239 of 273
Thread Starter 
Quote:
Originally Posted by Cantabrigian View Post

The point is the same. Those were two years with one very particular characteristic - very strong trends that (with the exception of Dec 08) that showed very little pullback from the trend. That tends to be the exception and generally backtested systems look overly good because of that sort of thing - it was smalls profitable most of the time and killed it in [fill in very specific circumstance that may or may not ever repeat]. It would be stupid to discount a model because it did well in those years and I'm not saying it's flawed because it did well then. All I'm saying is that it smells like overfitting. Out of sample tests should confirm or rebut that but as I understand it, it's a very trap to fall into.

All of our tests are out of sample. In sample testing is utterly meaningless, and would show us earning returns in the millions of percent annually. What kind of statistical novice does in-sample backtesting ?

Also, you're flatly wrong on the trend issue. Look at the first quarter of 09. It's nothing but extremely volatile chop with no discernible trend whatsoever -- and not coincidentally (because of the high volatility) that's actually where we make the majority of our returns in that year. Besides, we do over 30% in 2010 and 2011 anyway (and we're up 15% in six months this year, so on pace to do 30-40% annually) so while I'd love to do 100%+ every year, we understand that those sort of numbers simply aren't duplicatable outside of crisis years. It doesn't matter though -- we're robust enough that we can churn out huge alpha in years with inconsistent trends.
Quote:
Bolded part = classic Martingale strategy / fallacy. Whenever you lose, double up and you'll eventually win it back.
Money management is commandments 1 - 12 of trading. You can almost suck at everything else and make money if you're disciplined. Loses like that should scare away investors (who don't share you last name) but it should also worry you - a lot. What's your biggest drawdown? There isn't any reason that you couldn't see two or 1.5 of those back to back. The fact that it didn't happen on the backtest should provide you any comfort in that regard.

It's not a Martingale fallacy. I don't think you have actually read much about our model or our process, because you keep making huge assumption/errors in discussing it (like the 10x leverage thing). We never hold a position for longer than three days -- so putting in a "stop" after a 10% portfolio drawdown equates to ceasing trading for the year. Which is nonsense when your trading window is that small, and when your drawdowns never take longer than four months to fully recover, no matter their size.
Quote:
I'm interested - sometimes your descriptions make this seems like a trend following model but this makes it seem like a mean reversion one. Or maybe you switch between the two.
I was being flip - sure there are a million winning secrets. But there are also a million guys with a trillion dollars chasing each and every one of those secrets.

It's both. As I've mentioned many times in this thread, we have a stable of trading strategies that we dynamically switch between based on a voting system of machine learning algorithms. Some of the underlying trading strategies are mean reversion, and some are trend following.
Quote:
You're trading stocks / ETFs, not options right?

correct. Cash equities only.

Quote:
In that case you, what you think you're doing doesn't differ in any meaningful way from predicting price movements (1) because it's just as hard to guess the right strategy / market regime as it is to guess the direction of the market and (2) because the strategies that you switch between only profit off directional bets, you also need to predict direction.

(1) This is simply not true. I can tell you flat out that the data structure of the performance time-series that we use as input variables in the machine learning process have substantially more structure than price data alone. It's still non-parametric overall, but it's substantially less noisy and the strategies are built in such a way that they have intrinsic correlations, which also adds to the ease of prediction.

(2) The strategies predict direction, not the machine-learning systems. It's not a direct machine-learning/price-direction system.
post #240 of 273
Thread Starter 
Anyway, SALT was awesome. Made some really great contacts. Wouldn't be surprised if a 3-6 months down the road (for the due diligence process) some of these leads turned into a major allocation for us. Also got a chance to learn a lot more about potential service providers, which while much less sexy is still an important part of running the business. Also looks like we will almost definitely be offered the opportunity to be PMs of a substantial allocation at another firm while we wait for seed money for our own fund. Could be a good way to pay the bills and build pedigree/track record while we try to recruit capital for our own fund.

Plus, got to have a lot of fun. All the service providers treated me like the prettiest girl in the room because we haven't signed up with anyone yet and they all want to get in with the new launches. Lots of free dinners, clubs, and booze. Got to talk shop with some insanely smart people, and got some good ideas for potential tweaks to the system while learning more about the industry. Got to meet a lot of hot, smart finance girls. All in all, very productive and very fun, and not particularly expensive.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Business, Careers & Education
Styleforum › Forums › Culture › Business, Careers & Education › Starting an investment business