Last time: who gets paid. This time: how to play anyway without being the payer.
A week ago I showed you the receipts on the bot gold rush — three firms taking ~90% of the extracted value while the ad reads “passive income”. The point of that piece was not never build a bot. The point was never race the incumbents. Speed, venue proximity, block-ordering privileges — those games are over, and the entry fee is your account.
But a bot that isn’t racing anyone? That’s just software that executes a plan without waking up at 3am to check its phone, without revenge-trading a red morning, without “just this once” doubling down. Discipline, automated. That is a genuinely good reason to build one — maybe the only good reason.
So here’s the sane on-ramp. Short version, because the long version is where people go to hide from actually starting.
Step one: trade fake money like it’s real
You want Alpaca for this. It’s a brokerage built API-first — commission-free stocks and ETFs, and the part that matters for you: a free paper trading account. Real market data, real order mechanics, fake money. Sign up, and the dashboard hands you two API keys for a sandbox that behaves like the live thing.
The whole “hello world” is about ten lines:
from alpaca.trading.client import TradingClient
from alpaca.trading.requests import MarketOrderRequest
from alpaca.trading.enums import OrderSide, TimeInForce
client = TradingClient("YOUR_KEY", "YOUR_SECRET", paper=True)
order = MarketOrderRequest(
symbol="SPY", qty=1,
side=OrderSide.BUY, time_in_force=TimeInForce.DAY,
)
client.submit_order(order)That’s it. You now have a program that trades. It has no idea what to trade or how much — which makes it exactly as dangerous as most bots people run with real money.
Live in the sandbox for months, not days. Not because the code is hard — because the sandbox is where you find out your strategy’s real personality, and the tuition is free. The single most expensive sentence in retail quant is “backtest looked great, going live.”
Step two: the three rules that keep quants alive
Everything I’ve learned running automated strategies compresses to three rules. None of them are about picking entries.
1. Your backtest is lying to you. Always. The only question is how much. It skipped the spread you’d actually pay, the slippage on the fill, the outage during the one candle that mattered — and if you tweaked parameters until the curve looked good, you didn’t discover an edge, you memorized the past. A useful reflex: the better a backtest looks, the more suspicious you should get. Real edges look boring on paper. Fake ones look incredible.
2. Sizing beats signals. A mediocre signal with disciplined sizing survives. A brilliant signal bet too big dies on its first losing streak — and every strategy has a losing streak coming, including the good ones. Risk a small fixed fraction per trade, know your maximum tolerable drawdown before you start, and let the math of not-dying do its compounding thing. Amateurs ask “what should I buy?” Professionals ask “how much can I afford to be wrong?”
3. Every strategy has weather it dies in. Trend-followers bleed out in chop. Mean-reverters get steamrolled by trends. Nothing works everywhere, and the market changes weather without filing a notice. A bot that doesn’t know what regime it’s built for isn’t a strategy — it’s a coin with extra steps. You don’t need anything fancy to start: just write down, in one sentence, the market condition your bot expects. When that condition leaves, your bot should too.
Step three: hedge like an adult
Here’s the part almost every retail bot skips, and it’s the difference between a strategy and a donation.
Your bot is a position. If it’s long tech stocks, you are long tech — whatever the entry logic was. The market does not care how clever the signal that put you there was. So the adult question is never “is my bot winning?” It’s “what am I net exposed to right now?”
Two ideas cover most of what you need:
Hedge the exposure you didn’t mean to take. If your strategy picks stocks but you have no opinion on the whole market, you’re accidentally betting on the whole market anyway. Offsetting some of that broad exposure — an inverse ETF, an index short, or simply running smaller — turns “stock picker plus hidden index bet” into just the stock picker. That’s the whole trick behind market-neutral: keep the part you have an edge on, sell off the part you don’t.
Expect your hedge to cost money. A hedge is insurance. Insurance has a premium. If your hedge makes money most months, it isn’t a hedge — it’s a second strategy, and now you have two things you don’t understand. Judge a hedge by the drawdown it saved you from, not by its P&L. And know its limits: in a real crash, assets that “weren’t correlated” hold hands and jump together. Correlations go to one exactly when you need them not to. Size for that day, because that day is the one your account actually has to survive.
That’s it. That’s the adult version. Notice nothing in it is exotic — it’s mostly refusing to take bets you didn’t choose.
What I’m not going to tell you
I run bots. Several. You may have noticed this post contains exactly zero details about what they trade, what signals they watch, or how they’re hedged — and that’s not me being coy, it’s me making a point you should internalize: any edge specific enough to publish is dead the moment it’s published. If a strategy is printable, it’s priced. What survives contact with the market isn’t the signal — it’s the discipline wrapped around it: paper first, size small, know your regime, hedge the accident.
Get the boring parts right and the sandbox will tell you if you have anything worth scaling. Most people never find out, because they skipped straight to the part where they lose.
More of this coming — sizing math, regime detection, when to kill a strategy. The newsletter is where it lands first.
Mark trades his own book. Nothing here is financial advice — it’s plumbing advice for people who insist on the pipes.