Want to start a hedge fund, where to begin?

New btcdog
Oct 22 44 Comments

Background: 2 engineers with ML background. We started learning/building algorithmic trading strategies 5 months ago with the goal of launching a fund. Our algo is very conservative/opportunistic meaning the base premise is to conserve existing capital then jump on a buy/short when criteria is met.

Algo generates ~32% compounded annual returns in backtestes since 2004 and has done 14% since July (2019) in live trading. We’re continuously optimizing it but I think our v1 is ready for a $ raise. Next steps, advice?

comments

Want to comment? LOG IN or SIGN UP
TOP 44 Comments
  • Bloomberg / Eng nunna
    You think blind can help you on this. Go to wallstreetoasis.com
    Oct 22 2
    • New btcdog
      OP
      Just joined it, thanks!
      Oct 22
    • Microsoft presci
      See, blind actually helped.
      Oct 22
  • Salesforce benimeoff
    If you have a winning strategy why not go all in with your own money. The next step is to find investors that will give you more capital, a full hedge fund will require a good deal of overhead.

    A lot of strategies can break down when you try and do it at larger scale. Even jumping from 100k to a mill could change how your algo performs as your ability to make bigger trades at the same price point diminishes. So, grow slowly with small rounds.
    Oct 22 2
    • New btcdog
      OP
      Thanks for your reply.

      I put in my own capital into fund in July and more in August. Live trading is running on $200k as starting cap.
      Oct 22
    • Salesforce benimeoff
      Ya, I would get a lot more friends,
      Family, and angel money. The legal and compliance aspects create a lot of overhead for a hedge fund. For instance you could need a series 3/7/65 and/or be registered with sec and perform filings, etc...

      Not to be a hater, but 32% annual returns sounds way to good to be true. High chance you will hit a snag with scale or get smoked in market scenarios you havent worked into your model. Make sure you have plenty of stop loss code in place...
      Oct 22
  • New / Consultant
    jennatalia

    New Consultant

    PRE
    Spotify
    jennataliamore
    Get 1 yr of live trading under the algo. Make sure it's fully audited by a reputable firm.

    Then plenty of investment funds will throw money at you.
    Oct 22 7
    • New btcdog
      OP
      Slippage of <1% with the set price is tolerable.
      Oct 22
    • New / Consultant
      jennatalia

      New Consultant

      PRE
      Spotify
      jennataliamore
      See if you can tie your model into a Level II screen so it's possible to include available liquidity in the algorithm.

      1% slippage isn't too common, but happens more often than you'd expect from arguably the most efficient market in the world.

      Slippage + broker / exchange fees + data feeds generally eat away at alpha generated.

      If your 32% annualized returns still survives that, leverage up and go hard. Because you'd be sitting on a unicorn in the algorithmic trading space.
      Oct 22
    • New btcdog
      OP
      It’s currently ranked in the 87th percentile on Quantconnect. We haven’t even implemented/tested the list of optimizations for v2 yet. A lot to be done still!
      Oct 22
    • Waymo PdAX50
      +1 to overfitting being a hazard of backtesting. Ideally, hide half of the historical data from yourself permanently. I would only use it to compute validation metrics but never use it to drill down to specific examples that inform the development of your strategy.

      If you've fully built your algorithm looking at half of historical data, and you can show a good return on the other half, that's a better indicator of how you'll do in the longer run with live trading. Think of every time you've gotten an improvement that was inspired by looking at what your strategies did in a problematic case. It's easy to forget that you have had the advantage of replaying history over and over and learning from it but you won't have that advantage live.

      This is sincere advice and I'm not at all naysaying. Sounds awesome what you're doing.
      Oct 22
    • Citadel Securities / Eng bluetiger
      Yea any returns based on back testing are almost certainly due to over-fitting - it’s very difficult to understand how the algo truly performs without a considerably long live trading performance history
      Oct 22
  • Citadel Securities Xnrs10
    Don’t start a hedge fund. There are tons of firms that will seed you and set you up with infrastructure, etc without you having to do all the insanity of a regulated offering. Get a referral to Jump or Tower or one of a dozen other firms that will back you.
    Oct 22 2
    • New btcdog
      OP
      Could you elaborate? Just trying to get a better understanding of this option.
      Oct 22
    • Citadel Securities Xnrs10
      Network your way over to an intro at one of the couple dozen Chicago firms that will back traders. They will all have different options in terms of P&L split, what you vs they pay for, who takes first losses, etc. Depending on arrangement you will either become a partner in their LLP or you will be a 1099 but in either case you will own your IP.
      Another option that is similar is to get backed by a family office. FOs will have more patience for longer term holdings and worry less about you putting up super high Sharpes but you need to be able to accept more capital (not the $50MM+ necessary for HF economics to work but bigger than the few hundred thousand needed to make the prop shop route work)
      Oct 23
  • T-Mobile OneT
    In terms of setup you'll need an LP ideally in Delaware and LLC where ever you are.....you'll also need to register the LP as a foreign entity in your states otherwise banks won't work with it....after that you'll need to choose a Prime Broker....Goldman has a good offering but their minimum is 10M alternatively you can use Interactive Brokers as your prime broker if you at least have 1M to invest....

    Get the series 66/67 cleared, setup the structure that you work for the LLC and the LLC is the investment advisor to LP and charges a fee for the services provided....all the investment funds live with the LP / prime broker.

    If you and all your investors live in the same state then you don't need to register with SEC and use the blue sky rule otherwise register with SEC and I forgot what that form is called....

    If you don't have investors then your prime broker can also introduce you to the fund of fund and other first loss capital providers...

    Good luck and buy me a drink when you start making tonnes of dough
    Oct 22 0
  • Microsoft
    Tier 1

    Microsoft

    BIO
    #1 in Prestige
    Tier 1more
    Make sure you put a lot of effort into risk management. There should be multiple fail safes.

    Then, you should work on developing additional independent strategies.

    The market is already fairly efficient and it’s likely that any pattern you found hasn’t been exploited yet because it’s too small to be worth it to larger investors, and so it will stop working if you put in too much more capital.
    Oct 22 3
    • New btcdog
      OP
      Hey! :D

      The main algo is actually specially designed with risk management in mind as I didn’t want to lose own capital, so I went overboard with covering loses. In the backtests, we were actually doing 48-56% annually and monte carlo sims we had scored really good on the win/loss rate. Then I wanted to go overboard with protecting my cap and reduced take profit/draw down.

      It’s very modular, we made about 35 different algos in the past 5 months and took different aspects of the best ones and tested/optimized into the main one.
      Oct 22
    • Microsoft
      Tier 1

      Microsoft

      BIO
      #1 in Prestige
      Tier 1more
      I mean things like what if your VM fails?

      What if there’s an outage of that region of your hosting provider?

      What if the logic in your main algo fails for some reason, there should be a failsafe that cuts it off after losing too much money too quickly.
      Oct 22
    • New btcdog
      OP
      We’re using Azure batch services so down let us down!!

      But to also mention, we’re not making risky bets. For buying; the screener uses S&P 500 stocks that fit the criteria to buy/short. So extremely liquid companies, and we tend to hold them for 1-5 days and have a separate tool to avoid trading near earnings.
      Oct 22
  • Flowcast lulll
    Hedge funds are interested in strategies that scale to billions of dollars. They don’t waste time making 30k a year let alone 3m.
    Oct 22 2
    • New btcdog
      OP
      Hey there!

      We’re trading S&P 500 stocks that fit the criteria to buy/short. So “extremely” liquid companies, we’re also not allocating 100% into one company. Hypothetically speaking if all 500 companies met the criteria, we would hold all 500. In real time it’s diversified, so it’s holding 38 companies at this given moment. It wouldn’t have any problem scaling to $2-3 BB.
      Oct 22
    • Flowcast lulll
      I dont think you know what hedge funds are looking before but good luck, you’ll get there
      Oct 22
  • Intel /\/\
    Read flash boys
    Oct 22 0
  • PARC Tmkc
    Market dynamics change over time. Backtesting is good, but you will have to keep updating your strategy to maintain the current levels of safety and gains.
    Oct 22 3
    • New / Consultant
      jennatalia

      New Consultant

      PRE
      Spotify
      jennataliamore
      That's arguably the opposite.

      You'd want to find a strategy that works across market dynamics.

      Uptrend, downtrend, recession, high tax, low tax, interest rates. Then the alpha from the algorithm is predictable and easier to leverage up.
      Oct 22
    • New btcdog
      OP
      Exactly as you mentioned is the sentiment for our modeling. We come from math, computer engineering background. We both built bots for natural language processing at a SF startup for two years but traded equites on the side every day until it completely took over (5 months ago).
      Oct 22
    • PARC Tmkc
      I’ve been through similar. My experience was that it takes about 5-6 months before any new strategy gets ‘built into’ other HFT algos - which start anticipating your next trades - and make it harder to keep up the advantage.
      Oct 22
  • T-Mobile OneT
    Also SEC is only for stocks and options, if you plan on trading currency, community or futures then you'll also need to register with CFTC....
    Oct 22 0
  • Oracle y6wkzC
    Nice.
    Oct 22 0
  • PARC Tmkc
    Those are generic macro trends. Problem is future market dynamics could be different from all past situations in your training data- because it hasn’t happened yet.
    Oct 22 1
    • New / Consultant
      jennatalia

      New Consultant

      PRE
      Spotify
      jennataliamore
      Trying to model those is impossible for the exact reason you mentioned--they haven't happened yet.

      So a proper trading algorithm should take advantage of a universal market truth that persists through time, and through multiple changes in market dynamics.

      A good example would be the shift from fractional to decimal pricing. A lot of strategies died with the reduction in spread.
      Oct 22
  • VMware BurnedOut
    As people said start with your paperwork now. Have each month of trading audited and get your friends to put up more money. 700k - 1M. Also start looking for small family offices that manage money. They will not "give" you money, but may provide a managed account to trade in so they keep full control I. E. No withdraw time limits, minimums etc. After a while if you keep it up the doors will open. Money raising is as hard as the actual trading.
    Oct 22 0
  • T-Mobile OneT
    Having told you how to set up the fund, I'll tell from experience there is no free lunch!

    If your backrest is free of bias then your algo is taking a risk that is not obvious to you but is being compensated for that...

    Without divulging your algo do you know what is the source of your strategy's alpha? Because in poker there is a saying if you don't know who the Patsy is then it's probably you....
    Oct 22 0
  • PARC Tmkc
    The very act of executing such ‘universal market truth’ based trading changes the dynamics.

    Over time, other players / HFT algos will pick up on it and make it harder to profit from it.

    The only way to keep the high yield and safety will be to keep testing / updating / retraining.

    AFAIK, the only set-and-forget strategy based on a ‘universal market truth’ is found in Jack Bogle’s Little Book of Common Sense Investing - and it doesn’t yield anywhere close to 30% :)
    Oct 22 1
    • New / Consultant
      jennatalia

      New Consultant

      PRE
      Spotify
      jennataliamore
      Arbitrage.
      Oct 22
  • Amazon qyahbosjej
    is your startegy related to natural language processing techniques?
    Nov 14 0
  • Citadel / Eng
    BrutеForce

    Citadel Eng

    PRE
    EPAM Systems
    BrutеForcemore
    Backtest is good and all, but how does it perform on a new data?
    Oct 23 0
  • Roku eh ugh GC
    Your model is over fitted.
    Oct 22 0
  • Cisco Infensus
    Op what is your slippage model
    Oct 22 0
  • Amazon catowayne
    Backtest will wildly overfit. Make sure you’re not just testing on training data...
    Oct 22 0

Salary
Comparison

    Real time salary information from verified employees