Bitcoin and miner economics. Building an AI agent that trades the S&P 500 to stack more BTC. Every result posted. Weekly log below.
United States
Joined January 2023
- Tweets3.3K
- Following440
- Followers559
- Likes4.2K
Pinned Tweet
Q4 mining costs to support Bills claim. Last halving bear bottom was -25% AISC for $RIOT. I, the ever optimist, hoped we would hold 70-75k as that fit the narrative. Now that these levels have been broken, you must assume worst case, which would be right in @billfour range of 55-60k. If we get -25% below AISC of the cheapest miners we’re at 40k. I wouldn’t rule out anything in $BTC but I like bills range, especially after today.
$MARA $IBIT $IREN $ABTC $MSTR $HODL $COIN $WULF $BITF $HUT $CLSK
Fixed three bugs with our agentic trader this morning.
1. One position sat without a stop overnight. This was due to an order that got partially filled. Now a check every minute to confirm when a buy has completed so stops can be added.
2. Unfilled buys got dropped, now they retry.
3. GitHub scheduler ran late. Added an earlier start and a second check.
Some major bugs if you ask me so we will be keeping a close eye on the fixes.
Benjamin Pincock retweeted
This is the way. Stock donations to kids w inv accts helps to kill many societal problems w one clean stone. We know some will perform olympic-level mental gymnastics to try and spin this as bad but as someone who has looked into it, this is really powerful stuff. 👏
Excellent news - the engine and model fired up on Alpaca last night for the first time and successfully traded the entire day with no operator involvement. Objective achieved. Still trying to figure out how and what I want to share in terms of data. Might just end up giving most of it away.. in the name of science ofc. If we can make it an entire week with no firing issues I will start trade logs again and see if I can figure out a way to automate posts for some of our signals as they are happening live.
Some things are changing for the project we started running last week. We are moving from Robinhood to Alpaca for better automations in our trading. We want a fully automated agentic trader and Robinhood could not give that to us. As we are still in testing, I am going to pull back the posts and data entries to just once a week, every Friday so as to not spam my followers. If you are a builder and want more information on what I am doing send me a PM and let’s talk.
The model and engine in their current form are not ready to be deployed independently by an agent. There has been far too much operator involvement and that is not what we had in mind. We are taking this back to the drawing board and working out our automations. Stay tuned.
Benjamin Pincock retweeted
BREAKING: South Korea's stock market extends losses to over -8% on the day, triggering a level 1 circuit breaker.
Trading has been halted for 20 minutes amid the intense selling pressure.
A level 2 circuit breaker would occur next at a -15% drop.
Benjamin Pincock retweeted
Cavemen Issues Dire Warning That The Wheel Is Going To Eliminate Countless Jobs
Our back-test assumed one of our morning trades happens right at the 9:30 open. Our AI agent actually places it a bit later, on a schedule.
That's a hidden gap between model and reality, so we checked whether it matters. 🧵
The catch: the biggest gap between any two times is ~2 bp. A normal day swings ~80 bp.
For our own trade, the model-vs-agent timing gap was under 1 bp a day. The model stands; timing alone isn't an edge.
Bitcoin's wild swings are shrinking, halving after halving.
Typical yearly volatility by cycle: 152% → 94% → 78% → 65% → 43% so far.
Worst crashes: -93%, -85%, -84%, -77%, and -53% this cycle so far.
It now swings about as much as Nvidia. Not advice.
ALT Bitcoin volatility per halving cycle: 152% before the halvings (2010-12), 94% in 2012-16, 78% in 2016-20, 65% in 2020-24, 43% so far in 2024-now, and a dashed bar near 35% for 2028-32 if the trend holds. Worst drop each cycle: -93% (2011), -85% (2014-15), -84% (2018), -77% (2022), -53% so far (2026). Weeks with a 20%+ move: 18 in 2013, 0 in 2025. Peak vs halving price: 92x in the 2013 cycle, 2x in the 2025 cycle. Last 12 months: 44%, about what Nvidia swings.
At its heart, our engine isn't a trading bot. It's a savings plan.
Money goes in every month, no matter what the market does. The engine puts it to work. Its profits buy bitcoin. The bitcoin is never sold. 🧵
The engine's job is to fuel the bitcoin. Bitcoin's job is to be the long-term store we never touch.
Result: more return than SPY or the engine alone, a better Sharpe than SPY or bitcoin, a smaller worst drop than SPY. The cost: a deeper drop than the engine alone.