[pyrrhotech]

I've been building algorithmic trading models for the last 4+ years. After trading them successfully with my own capital for more than a year, I launched https://grizzlybulls.com as an alternative to the traditional hedge fund monetization path.

Since launching in January 2022, we've significantly outperformed the market with lower volatility and reduced max drawdown:

Model - Return - Max drawdown

S&P 500 (benchmark): +9.91% -27.56%

Platinum: +45.34% -16.48%

Gold: +39.53% -19.12%

Silver: +17.24% -22.96%

Bronze: +14.12% -23.93%

Vix Basic: +9.81% -24.23%

TA - Mean Reversion: +17.77% -19.92%

TA - Trend: +17.29% -24.98%

This is an unleveraged, apples to apples comparison. These are not high frequency trading models. Most of them only make a trade every 2-4 weeks on average. During long signals, the models are simply long the S&P 500 and during short signals, they go to cash. This can be implemented very tax efficiently by holding a core ETF long position that never gets sold and then selling S&P 500 futures (ES or MES) of equal value to the ETFs against the long position. This way your account will accumulate unrealized capital gains indefinitely and you'll only pay tax on the net result of successful hedging. The cherry on top is that the S&P 500 futures are section 1256 contracts that are taxed at 60% long term / 40% short term capital gains rates regardless of the duration they are held.

The models use a variety of indicators, many of them custom built. Most important are various VIX metrics (absolute level, VIX futures curve shape/slope, divergences against S&P 500 price, etc), trend-following TA metrics (MACD, EMV, etc), mean-reversion TA metrics (Bollinger Bands, CMO, etc), macroeconomic (unemployment, housing starts, leading composite), and monetary policy (yield curve inversion, equity risk premium, dot plot, etc). They've been backtested very cautiously to avoid overfitting.

[bbor]

Just popping in to say:

1. Love the name, not enough Pyrrhonists on hacker news these days! The OG.

2. Love the website, your design skills are killer. I hate that entire industry and even still my monkey brain went "oo I want to see the Euphoria index, sign up!"

3. This is kinda quintissential AI. Not to distract this thread from the valuable topic of non-trendy projects, but this is a great example of why we need to reclaim "AI" as a much more general term. I mean "algorithmic trading" could be a synonym for "human-like problem solving"...

[c_o_n_v_e_x]

Congrats on your success to date. I spent quite a bit of time putting together a trend system based on breakouts in futures markets. The system itself was nothing overly special. I purchased a few decades worth of futures data, created and backtested the system with Tradingblox.

The biggest problem was that the system really needed a minimum account of $1m USD so that each position wasnt too large and to get the diversification across different futures markets.

[m3kw9]

Real test is when you have a down market. Everyone one looks like a genius in a bull market

[WorkerBee28474]

The parent comment says they launched in Jan 2022. Jan 2022 to Jan 2023 was a down market.

[cosgrove]

This is nice to hear about. Can you tell me more about how your live results matched or diverged from your backtesting?

Did you list the returns of the commodities as a comparison, or are you trading those futures as well in the mix? (I know you only talked about ES/MES)

[pyrrhotech]

I've studied many systems over the years and never found any that matched or outperformed their backtests. So far our live results have hit between 1/4 and 3/4 of backtest performance depending on the model. Needless to say the high inflation and high interest rate market climate over the last two years hasn't been seen in the rest of the backtest period, but conditions are starting to normalize now.

Nevertheless, it would be prudent to expect any algorithmic trading model to underperform its backtest going forward, but there's enough leeway in the CAGR and max drawdown figures to underperform the backtest and still produce substantial alpha, especially for the more advanced models.

Right now the models are specialized to trade equities. I may develop new models that trade commodities in the future though.

[halfcat]

Is it correct to say you need a multiple of 50x the value of SPY to execute this strategy (if the minimum you can hedge with is 1x MES)?

[pyrrhotech]

To implement the strategy in the most tax efficient manner without leverage you would want to have an account worth 5 * (S&P 500 futures price). Today that would be about $26,375. MES uses a multiple of 5 while ES uses a multiple of 10.

However, with today's $0 commissions, if you aren't overly concerned about taxes, you can try out this strategy with as little as $500 and simply buy and sell one share of the ETF VOO on signal changes. Alternatively, if you have the risk appetite, you can get started with trading MES futures with less than $10k, though caution should always be warranted when using any amount of leverage.

[carabiner]

Have you retired yet?