AI Trading Advice Fact Check, Part 2: The Backtest Receipts

0:00Let's build an AI trading bot that can trade stocks while you s-

0:02Let's fact check this video.

0:05And we're pulling all of our data from the Yahoo Finance API

0:06which gives us access to all the trading data since the year 2000.

0:09If Yahoo Finance even thinks you're using their data like this, they are going to cook you.

0:16If you move in silence, don't tell anyone,

0:19have VPNs, then you'll probably get away with it.

0:22But also, why would you want to?

0:24Yahoo Finance data isn't all that good.

0:26It really only has open, high, low, close data and other sources of data like fundamental data,

0:32intraday data, options data, economic data.

0:36All of that's not really accessible on Yahoo Finance.

0:38I'm caching all my data locally so every run I don't have to

0:40re-pull the data and reprocess it all over again.

0:42This is great if you're using open, high, low, close data like he says in the video.

0:46But if you're using anything like quote data, intraday options data, alternative data sources,

0:50you're not going to be able to fit it on your computer unless you have a 20 terabyte drive.

0:54So you're going to need a new strategy.

0:56The data I'm pulling is from four funds.

0:58The S&P 500, the NASDAQ, international stocks and government bonds.

1:00And I built out four small programs that each work together to determine how much

1:02of the market we should own today.

1:03And the first program I kept pretty simple.

1:04And it's basically just an if statement with 25 years of data.

1:06So the way that it works is that if the NASDAQ is above the average of its last 200 days,

1:09it stays in and if not, it gets out completely.

1:11The thing that bothers me about this is this strategy is so simple that he knows

1:16you can go in Claude right now and backtest the exact strategy

1:20and compare it to any baseline you want.

1:23In fact, I sent the exact rules to my trading agent.

1:26If QQQ is above a 200-day average price, buy 100% of my buying power in QQQ.

1:31Otherwise sell all my positions.

1:34Let's see what my AI agent cooks up for me.

1:37And we can see the return.

1:39The green line is his strategy, right?

1:42The gray line is if you bought the S&P 500 and did nothing at all, you bought and held.

1:48And this green line is not only worse in every single way,

1:52but we also have to pay taxes, fees, and it's just not a good strategy.

1:57I mean, look at it.

1:59For program two, it runs once a month and it compares US stocks to international stocks

2:02over the past year and it buys whichever one of those are winning.

2:04And if both of them are losing, it stops everything in bonds and it just waits.

2:07Create a strategy that uses the international ETF and SPY ETF and buys which one is winning.

2:13If neither is, we go to bonds and wait.

2:16And even though my AI created different definitions of the word winning,

2:23not a single one of these portfolios even outperformed doing nothing at all.

2:31Program three was pretty simple and it basically just buys the dip.

2:32So if the market drops a ton for two days straight,

2:34but the trend is still up, it buys and it sells as soon as it bounces back.

2:37Buy the dip doesn't work.

2:39In many cases, you lose a whole bunch of money.

2:43Like why would you do this?

2:46When you can do this.

2:49He then goes on to create an ensemble of different portfolios,

2:52basically combining all of these weak trading strategies

2:57to see if it can create a profitable trading strategy.

3:00So if you want to see the results of that, like and follow for part two.

3:04But from what I've seen so far, I'm very much not convinced.

3:08It seems like you'll make a lot more money just doing nothing at all.

3:13Like why would you not just buy SPY?

3:16Let me know down in the comments.