Crypto Trading Bot Profits: What Actually Happened When I Ran One

🌐 한국어

Search "crypto trading bot profits" and you get two kinds of articles. One boasts about making some percentage a month; the other says it's all a scam. Neither is much help.

This is an account from someone who actually built a bot and ran it, describing what happened without promising returns. The short version up front: "unclear on the money, definitely learned a lot."

Everything Works in a Backtest

The first thing I built was a grid bot. Divide the price into fixed intervals, buy as it falls, sell as it rises — the simplest possible structure. Run it against historical data and the results looked good. Of course they did. Pretending not to know the future is harder in a backtest than people realize.

My first backtest had these problems:

• It assumed orders always filled at the price I wanted. In reality, a thin order book means you don't get filled where you asked.
• Fees were approximated or omitted entirely. For a high-frequency strategy, that's fatal.
• The most embarrassing one: I tested on a coin that had already gone up a lot. Validate on an asset you picked after knowing the outcome and anything looks good.

# My first backtest — don't do this

if price <= grid_buy_price:

    position += size            # assumes it just fills at the price I wanted

    cash -= grid_buy_price * size

# Fixed version — accounts for fees and slippage

if price <= grid_buy_price:

    fill = grid_buy_price * (1 + SLIPPAGE)   # in reality you buy slightly higher

    cost = fill * size * (1 + FEE_RATE)      # and the fee comes off too

    position += size

    cash -= cost

Adding fees and slippage and re-running dropped the equity curve noticeably. The more trades a strategy made, the bigger the drop. That's where one thing became clear: the biggest enemy of automated trading isn't the market, it's transaction costs.

What Live Trading Threw at Me First

I went live with a small amount. What I hit in week one had nothing to do with strategy.

• Minimum order size: I'd set the grid intervals tightly, so each order fell below the exchange's minimum and got rejected over and over.
• API rate limits: polling several markets on a short cycle earned me a 429. The bot did nothing at all during that window.
• State loss on restart: I restarted the process and the bot forgot its own position. It bought a quantity it already held. That one hurt the most.
• Helplessness in a trending market: a grid bot needs price to oscillate. When price runs one direction, you either keep getting filled deeper or keep missing entirely.

The state loss problem I eventually fixed by writing every position to a file and restoring on startup. Simple — and it should have been there from the beginning.

import json, os

STATE_FILE = "bot_state.json"

def save_state(positions, stats):

    tmp = STATE_FILE + ".tmp"

    with open(tmp, "w", encoding="utf-8") as f:

        json.dump({"positions": positions, "stats": stats},

                  f, ensure_ascii=False, indent=2)

    os.replace(tmp, STATE_FILE)      # existing file stays intact if we die mid-write

def load_state():

    if not os.path.exists(STATE_FILE):

        return {}, {}

    with open(STATE_FILE, encoding="utf-8") as f:

        d = json.load(f)

    return d.get("positions", {}), d.get("stats", {})

The reason for writing to a temp file and swapping with os.replace is so that a process death mid-save doesn't corrupt the existing state file. Small safeguards like this make a large difference live.

So Did It Make Money?

The honest answer: it varied by period, and overall it was not at a level where I could say "thanks to the bot."

In choppy, high-volatility ranges the grid bot ground out modest gains. In stretches where price fell continuously in one direction, it just kept averaging down and accumulating inventory. The paper loss grew, and eventually it erased the small gains that had built up before it.

That isn't a bug in the bot. It's how a grid strategy is designed to behave. The problem was that I ran it without understanding that characteristic well enough.

After several months of running it, here's the impression I'm left with.

• A bot is not the source of profit, it's a means of execution. Automate a strategy with no basis and you simply make baseless trades more often.
• Market conditions overwhelm strategy. The same bot does well some months and badly others. It is very easy to mistake a good month for skill.
• Short-run performance means close to nothing. Judge by a two-week return and you'll usually be wrong.

What I Got Anyway

Setting the money aside, building the bot was clearly worth it.

• Execution without emotion: with the rules encoded, there's no "just this once." That's exactly where discretionary trading always breaks down.
• A record exists: every decision and outcome accumulates in logs, so you can go back and find what went wrong. You never get this trading by hand.
• The skills stay: API authentication, retry handling, state management, running long-lived processes. That's useful engineering experience regardless of trading. Personally this was the biggest return.
• You start seeing the market structurally: order books, fills, fees, and slippage become concrete numbers. Vague intuition shrinks.

For Anyone Starting Out

I won't talk you out of it. But a few things are worth committing to.

• Set the goal as finishing, not profit. "A bot that runs three months without dying" is a good first target.
• Only use money you can afford to lose. That's practical advice, not a figure of speech. As the amount grows you stop being able to look at your code coldly.
• Discount backtest results by half. Even when you think you've handled fees, slippage, and selection bias, live will be worse.
• Avoid any bot or course that guarantees returns. If such a thing genuinely existed, there'd be no reason to sell it.

Automated trading is less a money-making machine and more a tool for converting your own judgment into something verifiable. What you do with that tool comes down to the person who built it.

This article records personal experience for educational and informational purposes and is not investment advice or a guarantee of returns. Cryptocurrency is extremely volatile, and automated trading carries loss risk up to and including total loss of principal. Past results do not guarantee future performance, and all investment decisions and consequences are your own.