Machine Learning powered trading for all

Dashboard
😍 Different model different color

Change colors on the dashboard to represent the different models. In the example above:

• Red is the TiREX model based on the xLSTM architecture.

• Black is the Cisco model that is based on the transformer architecture.

• Blue is based on a CNN that extracts features and passes them as residuals to an LSTM layer.

• Boxes are based on a computer vision model that predicts if the price is about to go up or down.

Dashboard
😍 Neater dashboard and dynamic port

• Every time you run the app, it spins up a unique port on your machine.

• The port can be located at the very bottom of the app.

• Links to the research papers and the code.

Broker
🥳 Work with any broker

When training the machine learning models, we now use data straight from your broker of choice.

8 TRADING BOT
🦾 8 Strategies to try out

We are working on different ways to use these models. We are welcome to suggestions. Meanwhile please test the 8 strategies we have made so far.

STRATEGY TRACKER
💹 Real-time market testing

Our mission is to build strategies using machine learning. As we backtest , we want to find the needle in the haystack for you. We kindly request that you join us in live testing for different brokers. Due to limited resources we are unable to test every broker available.

Until next time,
Wamaitha

PS:

You are receiving this email because you downloaded a trading bot from my kofi page or my gumroad page.

Howdy Trader!

I know, I know, its been way too long since I sent you an email. I have been freshly baking some new content that I think you will love!

Let me catch you up on what I have been up to.

I am sure you have seen the AI craze/hype/bubble thats been going on.

What if we could use AI to do something we are very good at but its very hard to put into code?

Let me explain with an example:

Hammer Candlestick Patterns

All three satisfy what it means to be a hammer candlestick. However, how little is the little in little to no upper wick ?

What does it mean mathematically to have a small body ? Is it 30% of the candle length? Maybe 20%?

What does a long wick even mean? What are we comparing it to in order to say hey! this wick is long and this one is medium.  

The rules are quite vague and the vagueness carries over when you want to create a program that identifies this things.

And thats where AI comes in!

During training, these models have seen all kinds of hammers , dojis, trends, you name it!

We can leverage this by asking AI to see the chart and draw these patterns for us.

Take this a step further, you can ask the AI to give a trading signal recommendation based on what they saw on the chart!

And thats what I have done with this Free trading bot.

It combines the accelerator oscillator and Gemini (AI model from Google) to build an AI assisted strategy.

Here is how its doing so far:

AI bounding boxes

Ai bounding boxes

I am very impressed with the performance on identifying the trends and the chart patterms but it has a long way to go with the candlesticks.

I know what your’re thinking…

but she just said we will use it for that!

before you come at me, bear in mind all that has been powered with this simple prompt:

Gemini Prompt

We have not given the model any more context than this.

For that, I think its doing pretty great!

With these AI models the adage garbage in garbage out is even more relevant! Poor prompt poor results.

On the next iteration I want to explore how we can better this prompt so that we can get the best of both worlds; using code and AI to build even more robust trading systems.

I have covered this in depth here:

As always you can try out the free bot here. Comes with a complete guide on how to setup everything no code needed!

Please give it a go and let me know how we can further improve it!

That’s all for today! Talk soon!

Howdy Trader!

Imagine you have $10 to invest in a lemonade stand.

Normally, with $10, you could buy a few lemons, some sugar, and maybe a couple of cups.

It's a small operation, but you can still sell some refreshing lemonade and make a little profit.

Your friend comes by one afternoon and really loves your Lemonade! In fact they believe in your lemonade-selling skills so much that they're willing to lend you a hand (or in this case, some cash).

Let's say your friend offers to lend you $90, bringing your total investment capital to $100.

This does two things for the business:

• Increased Buying Power: With $100, you can buy a lot more lemons, sugar, and cups. You could even upgrade to a bigger table or a fancy cooler to keep your lemonade icy cold.

• Potential for Bigger Profits: If the weather is hot and everyone's craving lemonade, you might sell out quickly and make a much higher profit than you could have with just your $10. In this scenario, leverage would have significantly amplified your gains.

If there are competitors in the area who are only operating on the $10 budget, you could snatch up some of their customers because your business now feels more focused and established.

Operating using the extra $90 can be likened to how leverage works. Essentially, leverage lets you control a much larger "position" in the lemonade market. The lemonade market in this case is the stock market.

Your broker can say “Hey, I believe in the Apple Stock as much as you do so here is some more cash for you to buy the stock!”.

If Apple goes to the moon, you stand to make more money than you would have made if you only bought $10 worth of stocks.

But there's a catch, just like there is with borrowing money:

• Risk of Bigger Losses: What if the weather turns out to be gloomy and nobody wants lemonade? With only $10 of your own money, you might just lose that and be out a few bucks. However, with leverage, you're still responsible for paying back the $90 you borrowed from your friend, even if you don't make any sales. In this case, leverage would have magnified your losses. The same applies to the stock, if Apple ends up collapsing, you would not only have lost your $10 but also the borrowed money. Which you still need to repay!

Your friend (broker) need to also put in some conditions in place to ensure you do not loose all their money AKA margin calls.

Learn from investing legends

Warren Buffett reads for 8 hours a day. What if you only have 5 minutes a day? Then, read Value Investor Daily. We scour the portfolios of top value investors and bring you all their best ideas.

Subscribe with one click.

Your friend (the broker) sets a rule that if the value of your lemonade stand (based on, say, the price of lemons and sugar) falls below a certain point, they need some reassurance you can still pay them back. This minimum value is called the maintenance margin.

Let's say the price of lemons unexpectedly doubles! Suddenly, your $100 worth of supplies might only be worth $60.

This means the value of your stand has dipped below the maintenance margin set by your friend. Here comes the margin call:

• Your friend calls you up (figuratively). They inform you that the value of your stand has dropped and needs to be brought back up to meet the minimum requirement.

• You have options. You can either:

  • Add more of your own money (your margin) to the stand. This is like adding more sugar and lemons to increase the value.

  • Sell some of your lemonade supplies (reduce your position). This might mean selling some cups or even your fancy cooler to get some cash back.

Leverage can be a great tool for small investors , allowing them to potentially make bigger profits.

However, it's important to remember that it also comes with the risk of bigger losses and margin calls.

Just like lemonade comes in different flavors, margin requirements can also vary depending on the asset you're trading and the broker you use. Here's how different margins are calculated:

• Percentage of the position value: This is the most common method. The broker might require a minimum margin that's a certain percentage of the total value of your leveraged position. For example, a 50% margin requirement on your $100 lemonade stand (position value) would mean you need to have at least $50 of your own money in the stand ($100 position value x 50% margin requirement = $50 minimum margin).

• Fixed amount per unit: With this method, the broker might set a fixed amount of money you need to put up for each unit of the asset you're trading. So, for example, they might require a $1 margin per cup of lemonade.

A word of caution! If you wouldn't overload your lemonade stand with ingredients you can't afford to pay back for, you shouldn't use more leverage than you're comfortable with in the stock market.

Howdy!

Look at the image below. What do you see?

The green are spikes

You can see two different strategies from this chart.

We can either “catch “ the spike(green) or “ride the wave” (red candles after the spike)

If you ride the wave, you would be selling on the Boom 1000 Index and if you catch the spikes you would buy.

The main challenge about spikes is that we never know when they will form.

To quote Deriv:

So how are we supposed to beat an algorithm that is pegged to nothing within the real world?

Do technical indicators even work?

Does the algorithm ‘think’ about support and resistance?

Does the algorithm change based on the current market supply and demand?

I attempted to answer the first question. Do technical indicators even work?

I used 5 different technical instruments in an attempt to make a profitable strategy.

The 5 being:

• Kaufman Adaptive Moving Average (AMA).

• Commodity Channel Index (CCI).

• Relative Strength Index(RSI).

• Stochastic Oscillator indicator.

• Support and resistance from a higher timeframe.

Something interesting happened when I used the AMA alone. The strategy was as follows:

And the backtest graph looked like this:

Straight up lost money for the entirety of the strategy.

I went back to the drawing board and arrived at the following:

The backtest results using this 5 looked like this:

backtest graph

The graph above shows the profit the strategy generated. It starts off being so profitable then a sharp dip happens.

On further investigation, the dip happens on February 18th 2020.

When you scroll on the Boom 1000 Index chart to this date, something interesting comes up.

We don’t have one but two “gaps”. The gaps look like this:

First gap

Second gap

You can tell the gaps are not Spikes. I reached out to the deriv support and they confirmed the gaps happen on an algorithm change.

This experiment does not exhaustively rule out technical analysis but it makes me wonder if we can actually use technical analysis to make money on Deriv.

I want to combine more indicators and see where that leads.

You can test out the bots on other assets and let me know if any comes up profitable.

Howdy howdy Trader!

If you have been in the trading space for a while you have probably come across countless people selling trading bots.

And not just that, they make sure to lock the bots to a specific account. This ensures you won’t resell the bot.

As you can imagine, this makes the business not all that passive for them.

They have to attend to the clients and compile the bots every time someone makes a purchase.

What if there was a way to make it all passive.

And by passive I mean, you won’t have to interact with the customer at all.

They come to your landing page, make the purchase, and automatically get sent the bot. All that for as low as $16 per month.

I have decided to attempt this kind of passive income so here is my plan:

1. Someone makes a purchase and gives their account information during purchase.

2. In the background, once the purchase is confirmed, have a system that auto compiles the bot and sends it to the users email.

This of course requires a trading bot to start with.

The Product

While I do not have a trading strategy to sell, I have decided to make a bot that covers all candlestick patterns.

The single, double and triple candlestick patterns. For example, when the doji happens or the three white soldiers occur, a notification and alert will be sent out to the user. Technically this is not a trading bot as it doesn’t do any trading but imagine swapping this for a working strategy.

Automation

The purchases will all be made on Stripe. Why Stripe? It’s easy to setup , has multiple payment options(even paypal) and has good developer support. If we ran into a technical problem we have access to great support from Stripe and other developers in the ecosystem.

On the checkout page, we ask the user for their email and three account numbers. This gives the user an option to add either demo or live accounts as they see fit.

Why three accounts you ask?

If I am being honest, it’s for selfish reasons.

I want to reduce as much customer support as possible. I can envision getting support requests like ‘I mistyped my account’ or ‘that account was closed’.

The user gets three shots to think long and hard about which accounts to use.

Once the payment has succeeded, Stripe notifies our backend of the payment and passes along the account numbers. Here is an example of what Stripe sends to our backend

{
  "object": {
    ...
    "custom_fields": [
      {
        "key": "accountnumberone",
        "label": {
          "custom": "Account Number One",
          "type": "custom"
        },
        "optional": false,
        "text": {
          "maximum_length": null,
          "minimum_length": null,
          "value": "963852741"
        },
        "type": "text"
      },
      {
        "key": "accountnumbertwo",
        "label": {
          "custom": "Account Number Two",
          "type": "custom"
        },
        "optional": false,
        "text": {
          "maximum_length": null,
          "minimum_length": null,
          "value": "4585417452"
        },
        "type": "text"
      },
      {
        "key": "accountnumberthree",
        "label": {
          "custom": "Account Number Three",
          "type": "custom"
        },
        "optional": false,
        "text": {
          "maximum_length": null,
          "minimum_length": null,
          "value": "6258784516"
        },
        "type": "text"
      }
    ],
   ...
    "customer_details": {
     ...
      "email": "wamaitha@peeppips.com",
      "name": "Wamaitha from YouTube",
      "phone": null,
      "tax_exempt": "none",
      "tax_ids": []
    },
    ...
}

The backend will be a Windows VPS. This is because compiling the bot requires a Windows environment.

Within the VPS, I will have three services. The first service waits for the payment event from Stripe. Once it receives the data, it compiles the trading bot and calls upon the second service which sends the email with the compiled bot to the client.

The email service also calls upon the third service that saves the data to a database in case we ever launch a new version of the product we can send the updated version to the clients.

Here is what the current iteration of the email looks like:

Automated email with the client bot

There is a programming concept called Idempotency.

This is a characteristic of a system, which guarantees that performing an operation multiple times will yield the same result as if it was executed only once.

Idempotency is the main objective with everything happening in the VPS. Theoretically, if the system is always at optimal performance we can have this setup once. Truly making it passive.

Maintenance costs ($16)

The upside about all this is we can easily run the system with as low as $16 per month. The $16 breaks down as follows:

This would be higher if you hired a customer support assistant. I will handle all the customer support requests myself.

For the $10 VPS, from my load testing, we can process at least 10 orders per second without throttling the VPS.

Unless you’re famous and launch the product to millions of people, this should be enough per month for anyone with a small - medium traffic.

If you get more traffic, you can always scale to a bigger server. You can afford it at that point!

As far as sending emails go, you can actually use some free or cheaper alternatives such as Amazon, Sendgrid or Brevo. My experience with all of them has been demoralising.

Turns out this industry is plagued with so many scammers that anytime you signup to these services and indicate that this is the kind of service you will be offering, the system locks you out. Hence, the use of the google workspace.

Don’t think that this is a ‘safer’ option to scam. Google has very advanced tech to detect scams, so you better make sure the service is legit or you will also loose the workspace subscription.

Technically this can be a $10 side hustle.

Perhaps the best part about this specific hustle is that the definition of a Doji today will still probably be the same 10 years from now. Unless new candlestick patterns are introduced the product remains ever green with no requirement to keep updating the source code.

Which brings us to the setup costs.

Setup Costs

Starting the hustle will probably be the most expensive part. Collecting payments requires a Stripe account.

Setting up the Stripe account requires you to have a verified business. Stripe also works with a select number of countries so if you’re in Africa like me, you will either have to setup a business in a Stripe supported country or use another vendor.

In my case, I setup the business in the UK and it cost around $180. This was enough to cover incorporation and setting up an address in the UK.

In retrospect I should probably have setup the LLC in Dubai or some tax haven. However, this would be relatively more expensive.

This arrangement also complicates taxes to some extent. Pay the UK government the corporate and digital tax and then pay income tax in my country.

The total break down for me was a total investment of $194.

The 1 month development time is because I am working on very many candlesticks. They are 68 in total as of this posts writing! This is like coding 68 different trading bots!

The development and automation costs for me are zero because I can do all the coding myself, however, if you were to hire a dev to do the coding, it could cost anywhere from $0 - infinity depending on how complicated the bot is, if the dev is part of the company etc.

Let’s use my rates as an example.

This would bring the costs on your part to $694.

The development costs will vary depending on the bot complexity and time taken. However, for easy math I have used the $300

If you priced the product at say $80, you would need 9 sales to break even.

Marketing

Setting up the system is actually the easy part. Without traffic its all useless. The adage ‘if you build it they will come’ doesn’t work.

I will cover my marketing plan on the next issue!

Thanks for reading!

Howdy Howdy!

I am going through this phase in my life where I feel like I can be more but the ‘more’ needs money.

In a quest to be more, I am making the jump.

I don’t want to bite more than I can chew, which is why I am starting with the lowest hanging fruit; API as a Service.

Let me explain with an analogy.

We have a King who has seven advisors. Anytime the King needs special counsel, he asks the relevant advisor. The king and the advisor can have this back and forth conversations when needed.

King and his advisors

The King in my case will be an AI model that we made a while ago and the advisors will be my potential customers.

I will host the AI mode on the cloud, then give people access to the model for a monthly subscription fee.

On a high level here is how the model works:

• Give the model candlestick data.

• The model uses the data to make a prediction.

This means that the candlestick data doesn’t necessarily come from MT5. It can come from MT4, MT5, other brokers etc.

I also don’t want to sell a dream like ‘this model will make you a millionaire’ or ‘this is the model of your dreams’

No!

The point is to use the model as an addition to what you are currently using.

While backtest results don’t guarantee future results, here are some backtest results over different assets.

This are results over a 5 year backtest.

META seems to have performed the best over that period.

So how does this work?

I have two end users in mind, someone who already knows how to code but don’t need to stress over how to host the model and someone who doesn’t know how to code completely.

On this issue I want to focus on the latter first.

The flow for the non techincal user involves 3 steps:

1. Making the subscription purchase on Stripe.

   

2. The user then gets an email like this

3. Lastly, when they download the bot from the instructions on the google doc, they use the API key to access the AI model.

   

4. That’s all!

You’re probably wondering what the £20 gets the user.

The user gets 10,000 requests per month. Here is a table showing how many requests per month each time frame has:

So let’s say you mainly trade on the 5 minute timeframe. If you put the AI bot on the 5 minute timeframe on assets that run 24/7 like the synthetic indices, you would send 8640 requests to my server.

This also means if you needed to also have the predictions on a higher timeframe like 4 hour timeframe, you could use the (10000 - 8640) = 1360 to use.

However, if you needed more than 10,000 requests per month I have made the pricing into tiers.

This means if you needed to have the bot on the 1Min timeframe all month, the second tier would suit you best.

The main reason for the cap on the number of requests is because I also need to pay for the underlying resources and make some small profit for the technical maintenance of the project.

This won’t be tied to any account, therefore, as long as you are yet to exhaust your limit, you can put the bot on as many accounts as you want.

Alright thats it! Happy trading!

Howdy Trader!

Today its all about money! We cover how you can make money on autopilot without needing to do much. (After the initial investment of course).

This is not one of those “Make money as you sleep by doing absolutely nothing!”

Sadly life doesn’t work like that. Something always has to give! That can be inform of money or time.

You cant make money from thin air. Consider even the most illegal businesses, they always need an initial investment of some sort.

Anyway, lets secure the bag, shall we?

If you’ve been around the trading world for long enough, you must have come across telegram and discord servers that give signals. Some paid some free.

The idea being that you can just use an experienced traders signals and make bank like they are.

Whether the signals are any good boils down to a lot of things.

Now flip the scenario.

You are a great trader. You’re strategy has been giving you net gains year on year and you think to yourself “How can I share this strategy with my close friends?”

Did you know that you can automate sending trading signals to discord or telegram?

All on autopilot, without you needing to do it.

Some caveats though! This works well for fully or semi automated strategies.

Let’s look at an example.

Remember the RSI + BBands bot?

(I have linked it at the end of this piece incase you want to test it out)

If you backtest the bot on the 15 minute timeframe, Boom 500 Index on Deriv, since 2019 you will notice the backtest actually makes money!

if you started with $1000 dollars you would have actually made a profit of $100+.

I know its not much but unlike the a million strategies out there, this one actually made money on a 3+ year backtest.

For this examples sake, lets say this is the holy grail strategy that mints money.

Instead of selling the bot itself as a one off charge, you could have the signals sent to discord.

Then make the discord access to the signals as a monthly subscription.

Whoa! whoa! whoa! All that sounds super!

However, you’re probably wondering how much it would actually cost.

Let’s take a closer look at the costs that would go into this. Let’s assume you only have a manual working strategy.

Let’s go over several ways you could execute this.

So here is the plan A.

You would need a couple of things for plan A. Plan A is perhaps the most expensive plan.

1. You would need to work with an MQL5 developer who would turn your manual trading strategy into a bot.

2. A VPS to host the bot so that it can run 24/7.

3. A developer who can help you host the code that would be interacting with discord or telegram.

The main cost here is the intial development of the strategy. Once you get over that, you will virtually be printing money.

You could save more if you found a dev who can do the MT5 bot programming and also has cloud hosting + can also use the discord or telegram APIs.

The main reason for separating the last segment from just putting everything on the VPS is because you would need to keep monitoring the server within the VPS.

If you’re not very technical and something happens, you might need to go back to the dev to help you set things up again. On top of which, you would need a bigger VPS that can handle the server on top of MT5.

If you still wanted to go this route, the costs would change as follows:

IKR! Better save the $6 and avoid future headaches!

In fact let’s call that plan B!

The cloud route makes it such that once its set, thats it! As long as the code can handle any errors without crashing, you’d be all set!

Lastly here is plan C:

Ditch the entire MT5 thing! Go straight to the source and use APIs. Deriv for example has an API. This means instead of needing to install MT5 on the VPS then load the bot on the VPS, you would be getting the Boom 500 Index data straight from Deriv.

What this also means, is that you would only need one dev who would code using the Deriv API and directly integrate the discord or telegram code all in one.

This can then be hosted on serverless! How serverless works is that Google cloud or AWS let you host the code on their servers and only charge you when the code runs.

If the code is not running, you pay nothing! Nada! Zilch!

You would only pay for what you use!

The new costs would then be as follows:

This leaves out one crucial part of the equation. How do you keep track of who renewed the subscription and who did not?

Alright! Alright! Let’s keep going on the money stuff!

In order to keep tabs on who is still subscribed and who is not, you can have a couple of options.

Option one:

Literally keep tabs on the daily on where the subscriptions are at! One workflow could be using a payment processor like Stripe or PayPal and activate the payments as a subscription.

The Stripe and PayPal platforms actually manage the subscriptions for you. While that sorts out the auto payment processing, you would need to manually kick out users who fail to renew the subscriptions.

Option two:

Use PayPal or stripe to manage customers then have a dev automate the member management on auto pilot.

This option would add up to the costs you will have incurred for the trading bots. Nonetheless, chances are, the same dev who will setup the cloud hosting will be able to setup the members management.

The additional costs now would come from increased scope of work.

Option three:

Use a service like Whop that lets you sell discord memberships. They handle all the payment processing and also kicking out the members.

There’s a catch of course!

If your memberships is $10 per month, they get 30% of that.

Perhaps a question to ask “Would it be cheaper to have this setup by a dev as a one off project? Or would it be cheaper if I made 70% of all my memberships but have none of the headaches?”.

I will let you pick that bone!

That’s all for today! Happy hunting!

Howdy trader!

I have spent the last couple of days attempting to improve the winrate of our AI model.

The first idea I had was to use the EMA 200 as a trend determinant.

If the price was above the EMA, the assumption would be that we are on an up trend.

If the price was below the EMA , we would be on a down trend.

I implemented the EMA condition on python and did a backtest for all timeframes.

The entire backtest took around 6 hours.

Here are the results:

The backtest on the minutes timeframe was fruitless. The data did not have enough variation when plotted on the lower timeframes. For comparison, here is the plot on a 1 day timeframe:

Plot on a 1 hour timeframe:

You can tell the 1 hour is so compact the model is unable to make out the trend properly.

There are a few things we can try to make this work on the lower timeframes:

1. Change the normalisation technique. A probable cause of the short candlesticks is that the price difference between two candles is so little that when its normalised, they almost fall on the same range.

2. Avoid normalisation all together.

3. Use less bars on a chart.

4. Use less bars + avoiding normalisation.

For now, I will first focus on trying out the model on the live markets. I will only focus on the timeframes that had more than 50% winrate.

Stay tuned for that!

Howdy Trader!

What if there was an AI-powered model that would inform you when a spike is about to happen?

What if I also told you it’s free of charge?

Would you believe me?

Whether you believe me or not, the AI has landed!

More specifically, we are dropping it here for you to try out.

Keep reading!

Present for you

The model is mainly used to predict trends on a live video feed of charts.

It’s supposed to work more or less like your eyes.

It “sees” the charts and tells you whether the trend is about to be on an uptrend or a downtrend.

I have modified the model to work with images instead.

After all, videos are images stitched together!

Let me explain with an example:

Assuming we want the AI to work on the 5-minute timeframe, We get the OHLC of the last 100 bars.

We then use the OHLC to draw their values on a blank canvas

We then feed the canvas to the model so that we can get a prediction.

Depending on the prediction, we can take action as needed.

Here is a sample of what the predictions look like:

I know it doesn’t look like these are candlesticks but they are! I have normalised them to values between 0 and 1 then scaled them up to fit the canvas.

Here is a gif of what the predicted images stitched together look like:

This smaller size also speeds up computation time.

The model is not perfect but can be a good signal for trend reversals.

On backtest, the model has a 54% win rate.

I took it a step further and added the EMA 200 to determine the trend. The strategy now is:

Hey friend!

Another week, another finance paper!

Imagine having a smart buddy, like Siri or Alexa, where you can ask money questions, like "What will Apple's stock price be next week?" or "How's my investment portfolio doing?"

We're diving into a study about how well smart assistants like ChatGPT or Bard answer these financial questions.

These assistants are great at chatting like humans, but can they handle finance questions without being trained in finance?

The research tests them on various finance tasks to see if they can figure things out on their own.

It looks into whether these assistants can manage financial inquiries based on their existing knowledge and if they're reliable for making important money decisions.

They compared ChatGPT to others, finding some performance gaps, especially with private data, suggesting some issues with data quality.

Despite these challenges, ChatGPT still does a good job at financial tasks without any specific finance training, showing its potential usefulness.

Excitingly, new models like BloombergGPT, trained on financial data, are now available, making it an exciting time in the world of finance and AI.

However, we are still faced with the challenge of the models being closed-sourced. For example, BloombergGPT is quite expensive for the common man to afford.

So where does this leave you as a retail trader? The big sharks will keep having access to these models which may be giving them an edge in the market.

I don’t think technical analysis alone is enough.

You need fundamentals, keep tabs on the news, keep tabs on social media, etc. For example, Elon can tweet anything and that would change the price of any of the stocks of his companies.

The paper can be found here.

That’s all for today. Happy trading!

Wamaitha Nyamu

Say hi 👋 on Instagram, TikTok , YouTube or LinkedIn

👩‍💻 Work with me on your trading bot here.

Another week, another free bot!

Imagine you're dropping a ball, and it bounces back up after hitting the floor. In trading, the "floor" is called support.

It's the price level where a lot of buyers think, "Hey, this is a good deal!" and start buying, stopping the price from falling further.

Now think of the ball hitting the ceiling and coming back down. That's resistance in trading.

It's where the price of an asset goes up, but then a lot of sellers think, "This is as high as it goes!" and start selling, stopping the price from rising more.

Why Does It Matter?

Understanding support and resistance is super important because it helps you as a trader make smarter decisions.

By looking at past prices, traders can guess where the support and resistance levels might be. This helps you figure out good times to buy or sell.

One tricky part is that seeing these levels isn't always clear-cut. It can be subjective, meaning different traders might see different support and resistance levels on the same chart.

Plus, when you’re using bots to trade for you, they can't "see" these levels as we do. They use past high and low prices to figure it out.

When programming such a trading bot, one key thing is deciding how far back in time to look on the chart.

For example, looking back 50 bars on a 4-hour chart shows different support and resistance levels than looking back 100 or 200 bars.

That’s not all!

The timeframe you choose (like 1 minute, 4 hours, or 1 day) changes how far back you're looking, which affects the support and resistance levels the bot uses.

While setting up this bot, we had to tell it a few things, like how many bars back to look for support and resistance, what timeframe to consider, and other details like how much to trade and where to set stop losses.

Finding the perfect settings is still a bit of an adventure for now.

Wamaitha Nyamu

Say hi 👋 on Instagram, TikTok , YouTube or LinkedIn

👩‍💻 Work with me on your trading bot here.

Howdy Trader!

Consider having a smart virtual assistant akin to Siri or Alexa, and you wish to pose financial inquiries like "What will Apple's stock price be next week?" or "Could you assess my investment portfolio?"

Today we look at a financial paper that covers how well large language models (LLMs) like ChatGPT or Bard perform when asked specific financial questions.

These LLMs possess remarkable proficiency in comprehending and generating human-like text.

Nonetheless, can they provide precise and dependable responses to such financial queries devoid of any prior exposure to financial data?

The paper conducts experiments involving these LLMs across various financial tasks to ascertain their performance in a "zero-shot" scenario, indicating that they haven't received specific training for financial applications.

 Researchers make comparisons to determine whether these models can truly excel in the financial realm or if they fall short.

The paper explores how proficient virtual assistants, empowered by advanced language models, are at handling financial inquiries without any prior knowledge of finance.

It also evaluates whether they can be relied upon for crucial financial decisions. The findings of this study offer insights into the capabilities and limitations of these AI models within the domain of finance.

The research assesses ChatGPT's zero-shot capabilities against other open-source LLMs and fine-tuned PLMs in finance.

However, a performance gap is noted, especially with non-public datasets, suggesting data quality issues.

Additionally, while generative LLMs offer advantages for annotating large text volumes, they require significantly more time for labeling.

Findings reveal that although fine-tuned Personal Language Models(PLMs) typically excel, ChatGPT still performs remarkably well without specific training, showcasing its potential for financial tasks.

Since the paper was published more LLMs have come out including BloombergGPT which has been specifically trained on financial data.

Very exciting times to be in!

If you loved today’s post, you will enjoy reading Simplify Wall Street Newsletter. Subscribe below

Sponsored

Simplify Wall Street

Logic-Driven Investment Alerts for Independent Investors.

Subscribe

The paper can be found here.

That’s all for today. Happy trading!

Thanks for reading! Reply anytime!

Howdy Trader! 👋🏾

Another week, another free trading bot!

Something good is cooking this year! Looks like Kismet is on our side! Another bot that made money on backtest! 💵💵. Let’s dive right in!

This bot also challenged me to use a different approach to risk management. Previously I would let you choose the number of pips for the stoploass and the take profit you wanted to use. For this bot, we are using the current account balance and the lot size you choose to determine the number of pips to set the stop loss and take profit. Let’s do some math, shall we?

Let’s say your balance is $10,000 and the price on the EURUSD chart is 1.0801, and you want to risk 1% on every trade, that would mean every trade you would be risking $100.

For the standard lot, 1 pip typically equals $10. So the price would need to move 10 pips.

In a Mini Lot (0.1 lot): 1 pip equals $1. Therefore the price would need to move 100pips.

For the Micro lot, 1 pip equals $0.01 , the price would need to move 1000 pips.

For the Nano lot, 1 pip equals $0.01, price would need to move 10,000 pips.

The strategy works as follows:👇🏽

For the Buy setup the MACD crosses above the signal line while both lines are below the zero level we ensure ADX is above or around 40.

For the sell, we ensure the MACD crossed below the signal line and both are above zero and ensure ADX is above 40.

That’s all for this week!

Before you go, please help me answer the poll below 👇🏽.

Wamaitha Nyamu

Say hi 👋 on Instagram, TikTok , YouTube or LinkedIn

👩‍💻 Work with me on your trading bot here.

Howdy trader!

Another week, another free bot! The bot of the week is a special one because, for the first time, I have backtested a bot that made money. Not only that, I did not have to tweak the original strategy to make it work! Let’s dive right in!

The bot works with two indicators the RSI and the Bollinger bands! Here is how it works:

Short Position Strategy:

When aiming for a short position, keep a keen eye on the RSI. It should be above 70, indicating that the asset is overbought. But that's not all; watch for a crucial moment - a break below the upper Bollinger Band. As soon as you witness a candle closing below this upper band, it's your signal to enter a sell position.

Stop Loss (SL): Place your stop-loss order if the price hits the upper Bollinger Band.

Take Profit (TP): Look to take your profit when the price reaches the lower Bollinger Band.

Long Position Strategy:

For long positions, focus on the RSI being below 30, indicating oversold conditions. Now, pay attention to another critical breakthrough - a break above the lower Bollinger Band. When a candle closes above this lower band, seize the opportunity to enter a buy position.

Stop Loss (SL): Set your stop-loss order if the price touches the lower Bollinger Band.

Take Profit (TP): Consider taking your profit when the price hits the upper Bollinger Band.

I did backtest on several assets, however the most profitable of them all has been the Boom 500. During the backtest period of 2 years (1/1/2022 - 22/1/2024) with an initial deposit of $1000, the bot made $555+ in profit! While the period does seem too long just to make 50% profit, this goes to show how hard it is to have a strategy that works well over a long time. However, as always, you can try out the bot for yourself on different timeframes and assets, you might stumble upon assets that would ideally make you more money!

Here are the bot inputs:

You can play around with them on any asset. I have coded the bot to work on any timeframe and any assets. Enjoy!

Wamaitha Nyamu

Say hi 👋 on Instagram, TikTok , YouTube or LinkedIn

👩‍💻 Work with me on your trading bot here.

Hello friend!

Greetings for the season! Hope you had a great holiday!

It’s a new year and we start the year with a bang! A free trading bot/indicator.

In this week’s experiment, we are implementing this video.

How the Bot Works:

The bot focuses on two EMAs- the EMA 9 and EMA 20. The bot relies on the price regressing to the EMA9 and using the EMA 20 as the SL.

BUY SETUP

When the price goes below the EMA9 and EMA20, we start buying. The ‘hope’ or ‘assumption’ is that the price will revert at some point and make money with the buy trades.

SELL SETUP

When the price goes above the EMA9 and EMA20, we start selling. The ‘hope’ or ‘assumption’ is that the price will revert at some point and make money with the buy trades.

Coding this strategy raised a few concerns on my end:

1. The strategy doesn’t implement any form of risk management. NO SL or TP was implemented in the strategy.

2. There was also an excessive amount of trades taken. I don’t see how this would work with a small account.

3. If the price was on trend, this would easily blow the account.

Overall, I made a few changes to try and add some form of risk management.

Modifications to the Strategy:

I only implemented the sell scenario and made 5 versions of the same strategy.

1. Modification 1 - Sell Limit: Instead of using the market execution of a sell, we use a SellLimit order with a flexible value in pips. This tries to mirror the original working of the bot.

   • Backtest Results:

     

2. Modification 2 - TP and SL: Adds a Take Profit (TP) and a Stop Loss (SL) at a risk-reward ratio of 1:2

   • Backtest Results:

     

3. Modification 3 - Volume Distance: Considers the distance between volume and EMA20.

   • 

4. Modification 4 - Asian Market: Trades only during the Asian market and considers the level of the 200EMA.

   • Backtest Results:

     

Bot Inputs

• MAOnePeriod : The first EMA, default to 9.

• MATwoPeriod : The secondEMA, default to 20.

• MAOneAppliedPrice : The first EMA applied price, default to using the CLOSE.

• MATwoAppliedPrice : The second EMA applied price, default to using the CLOSE.

• MAOneMode : The first EMA mode, default to using the Exponential.

• MATwoMode : The second EMA applied price, default to using Exponential.

• MAOneShift : The first EMA shift, default to using zero.

• MATwoShift : The first EMA shift, default to using zero.

• distanceFromMAOneinPips : This sets a buffer in terms of pips to how far you need the price to move to for the bot to start placing sell limits.

• incrementalOpenTradesEveryXPips : This is the difference in pips between the sell limit orders placed .

• theMagicNumber : this is a unique bot identifier

• theLotSize : the volume

• maximumSellStopOrdersToBeOpen: maximum number of sell limit orders to be open at any one time

• tpPips : take profit in pips

• takeTrades : when set to true, the bot will place a sell limit order. When set to false the bot will not place a sell limit order.

• notifications : when set to true the bot will send notifications when a sell limit order is placed.

What's Next?

• All 4 modifications will be tested on a Virtual Private Server (VPS) for a week on a demo account. If you're intrigued and want to follow the strategy, join our Discord channel and look for the bot1 channel.

• Have a strategy you need to be coded? Work with me here.

Important Note:

• Please refrain from trading these bots live; consider using them as signals.

• Supervision is crucial, especially for this strategy, so all bots have the option to turn off trading by default.

Wamaitha Nyamu

Say hi 👋 on Instagram, TikTok , YouTube or LinkedIn

👩‍💻 Work with me on your trading bot here.

Hello - It’s Wamaitha 👋🏾

In case you missed it:

• On Monday we started a project on using genetic algorithms to optimise the MACD parameters. The default parameters are 12,26 and 9.

• On Tuesday, we dropped a free bot using MACD and the 200 EMA with a 1:2 risk-reward ratio.

• Yesterday we looked at optimizing MACD parameters using Genetic Algorithms

• Today we look at FinGPT. A language model used in the world of finance.

Let’s dive in.

In a remarkable stride forward in the realm of artificial intelligence (AI), researchers have unveiled FinGPT, a groundbreaking AI model designed to revolutionize the world of finance. This innovative project aims to empower AI with the skills necessary to execute diverse financial tasks, making it accessible and efficient for many users. We'll delve into the significance of this advancement, the training process behind FinGPT, and its potential implications for financial professionals and enthusiasts alike.

The Significance of Sufficient Training Data

Imagine teaching a language model about money; it's akin to equipping it with the knowledge of a new language. Just as individuals need exposure to a new language to gain proficiency, AI models require ample training data to grasp the intricacies of finance. However, the challenge lies in gathering and structuring this data, often hindering the retrieval of crucial financial information.

The Birth of FinGPT

In response to this challenge, researchers developed FinGPT, a domain-specific language model meticulously crafted for comprehending and generating financial content. This AI model addresses the data collection hurdle, making it easier for individuals and organizations to harness the power of AI for diverse financial tasks without the need to amass extensive datasets or train their own models.

FinGPT's Two-Step Training Process

The development of FinGPT consists of two distinct phases. First, the language model is exposed to a vast corpus of online textual material to acquire foundational linguistic knowledge. Subsequently, the model is refined using a carefully curated collection of financial facts. This two-step training process equips FinGPT with an extensive understanding of language and finance.

Empowering Users and Professionals

The introduction of FinGPT simplifies the use of advanced AI technology, making it accessible to a broader audience for executing various financial tasks with ease. Users can benefit from FinGPT's capabilities to explain financial news, dissect market trends, and provide valuable insights into investment approaches. Professionals in the financial sector can utilize this tool for tasks such as generating financial reports, analyzing market data, and formulating recommendations for investment strategies.

Responsible AI Usage

In the quest for progress, it's crucial to emphasize responsible AI usage. The authors of this research emphasize the importance of deploying FinGPT responsibly. This potent tool must adhere to best practices to prevent disseminating inaccurate information or biased forecasts. To address this, the authors propose recommendations to enhance control and mitigate bias in the outputs of FinGPT, especially in the generation of financial literature.

Conclusion

In conclusion, FinGPT marks a significant milestone in the field of artificial intelligence, offering a specialized solution for comprehending and generating financial language. This innovation can potentially streamline financial tasks for both consumers and professionals. However, it also underscores the importance of responsible AI usage to ensure a positive impact on society and to guard against the spread of misleading information in the financial domain. With FinGPT, the future of AI in finance is brighter than ever, promising efficiency, accessibility, and ethical deployment.

The paper can be found here.

That’s all for today. Happy trading!

Wamaitha Nyamu

Say hi 👋 on Instagram, TikTok , YouTube or LinkedIn

👩‍💻 Work with me on your trading bot here.

Hello Friend!

In case you missed it:

• On Monday we started a project on using genetic algorithms to optimise the MACD parameters. The default parameters are 12,26 and 9.

• Yesterday, we dropped a free bot using MACD and the 200 EMA with a 1:2 risk-reward ratio.

• Today we look into optimising the MACD Parameters using GA.

Genetic Algorithms Steps

The experiment started off with an initial population of 100. A population is randomly initialized with MACD settings. Typically looks something like this:

1. [12,3,56] 
2. [43,2,55] 
3. [66,87,12] 
4. [98,77,104]
5. [24,67,88]
.
.
.
100. [1,44,66]

The second step was retrieving data from MT5 to use for backtesting. The EURUSD on the 5-minute timeframe was used, going back the last year. The win rate was used as the determining factor of a good combination. For example:

1. [12,3,56]    ---> 20% winrate
2. [43,2,55]    ---> 24% winrate
3. [66,87,12]   ---> 12% winrate
4. [98,77,104]  ---> 34% winrate
5. [24,67,88]   ---> 56% winrate
.
.
.
100. [1,44,66]  ---> 10% winrate

In the example above, the fifth set of settings would be considered the best combination. The win rates are then sorted from the highest to the lowest. From these winners, we select the best 20% ie 1 through 20. These become the “parents”. Since every population requires 100 participants, 80% is obtained through reproduction. Reproduction can be flipping the fast period of two parents to create a new “offspring”. Emulates how chromosomes from parents form an offspring.

Some of the offspring then go through mutation. Mutation occurs by randomly flipping an element on an offspring. Assume the offspring after reproduction between 4 and 5 yields this:

[98,67,88]

This offspring can undergo mutation by randomly changing an element to this:

[98,2,88]

Once we have a new set of population, we run a backtest again and get the best individuals. This is done for 50 Generations. In the end, we have the very best individual after going through 50 generations. Now on to the results.

Before we can get to the results. Some interesting challenges:

1. The genetic algorithm code was written in Python programming therefore we made the MACD calculations in Pythoon which might differ from what MT5 does. The difference is not much but it might have affected the results.

2. Backtest results did not account for slippage or server delays.

3. After running the algorithm for 3 days, I could not get past a 51% win rate. Meaning I might have needed to either keep the random range initialization function higher or lower.

4. The algorithm was very time and resource-intensive.

Results

After two days of running the algorithm, here are the winners

{
  'fast_ema_period': 94,
  'slow_ema_period': 182, 
  'signal_period': 170
} 

Winrate: 51.28205128205128 % 

{
  'fast_ema_period': 94,
  'slow_ema_period': 182, 
  'signal_period': 170
} 

Winrate: 51.42857142857142 %  

I’m yet to test this on MT5 however, you can use the free bot I sent out yesterday to test these win-rates on MT5.

Food for thought:

1. Would filtering trade times to only work during Asian market hours make this a better strategy? Considering the high volume that is traded during this time.

2. Would implementing a trend indicator to filter out the raging markets improve the win rate?

3. Which other indicator can we optimize?

4. Would we have a higher win rate on another currency pair or timeframe?

You can download yesterday’s free bot from here. The instructions for loading the bot on a VPS are also included in the downloaded zip file.

That’s all for today! Happy trading!

Wamaitha Nyamu

Say hi 👋 on Instagram, TikTok , YouTube or LinkedIn

👩‍💻 Work with me on your trading bot here.

Hey Friend!

Free bot of the week works as follows:

The video source on YT has over 1.3M views.

Backtest Results:

On Boom 1000, the bot did not make profits, however, on the trades where money was made, the risk-to-reward ratio was working quite well

Backtest results

The strategy needs a way to filter false signals even further. Some observations on Boom 1000:

• The trailing SL only works when the initial lot size is more than the minimum lot for Boom 1000. For example, when a trade is placed with an initial lot size of 0.2, we cannot close the partial 25% of the trade since Deriv doesn’t accept any volume below 0.2. I’ve made the lot size flexible on the Bot Input tab. Feel free to experiment!

• The strategy doesn’t do well on raging markets despite having the 200MA to filter out trends.

You can download the free bot here. The instructions for loading the bot on a VPS are also included in the downloaded zip file.

Wamaitha Nyamu

Say hi 👋 on Instagram, TikTok , YouTube or LinkedIn

👩‍💻 Work with me on your trading bot here.

/

Hello Friend!

MACD settings can be adjusted but default to 12,26,9

To understand what Genetic Algorithms(GA) are, let’s use an analogy:

Now this applied to the MACD:

For this experiment, we will closely emulate this paper that optimizes MACD parameters for Soybean Futures. The paper concludes that the new parameters improved the profits generated.

Wiles, Phoebe & Enke, David. (2015). Optimizing MACD Parameters via Genetic Algorithms for Soybean Futures. Procedia Computer Science. 61. 85-91. 10.1016/j.procs.2015.09.157.

The fitness function will implement the Signal Line Crossover.

• Bullish Crossover: When the MACD Line crosses above the Signal Line, it's a signal that the stock may be in an upward trend, suggesting a potential buying opportunity.

• Bearish Crossover: Conversely, when the MACD Line crosses below the Signal Line, it indicates a possible downward trend, suggesting a potential selling opportunity.

Thanks for reading! Reply anytime!