|
no
経験
|
0
製品
|
0
デモバージョン
|
|
0
ジョブ
|
1
シグナル
|
0
購読者
|
Mobile Engineer
において
Sydney
My main focus is fully automated trading. I don’t want to introduce too much manual input, but I still need to guide the model to understand risk and overall market behavior properly.
I’ve already implemented a trading signal, but it’s not profitable yet. Starting from March 2026, I’ll begin experimenting with a reinforcement learning agent to see if it can better exploit the market using my handcrafted signals.
If there’s truly no edge in the system, then that’s where the story ends. Well, I am seeing some strong evidences that the edge does exist, but it also means I have more things to work on.
https://www.mql5.com/en/signals/2361594 - F.U.N.N.Y
I’ve already implemented a trading signal, but it’s not profitable yet. Starting from March 2026, I’ll begin experimenting with a reinforcement learning agent to see if it can better exploit the market using my handcrafted signals.
If there’s truly no edge in the system, then that’s where the story ends. Well, I am seeing some strong evidences that the edge does exist, but it also means I have more things to work on.
https://www.mql5.com/en/signals/2361594 - F.U.N.N.Y
友達
プロファイルまたはユーザー検索を経由して友達を追加できます。追加するとオンラインステータスがわかるようになります。
Yiheng Quan
After lots of backtesting and finetuning, the latest version is deployed. I have implemented an order splitter, so instead of 0.05 lot of SpotCrude, it will become 5 x 0.01 lot. This is preparing when the fund scales up. Max deposit load will drop to around 40% to reduce drawdown, but still expect 20 - 40% drawdown at some point.
Yiheng Quan
Trailing + flipping didn't seem to work, the backtesting had some issues in the end. Overall, SpotCrude is the only valid ticker with over 4% profit on average every month. There are foundamental issues with the current system, but I will keep it open until the new version is done. Another new journey, but this time, I know what to do, no longer searching all the possibilities like back in 2024.
Yiheng Quan
Trailing stop is now applied, this helps significantly in the backtesting, one of the top two boosters.
Yiheng Quan
SpotCrude will now keep the double lot and having more priority, other tickers will revert back to regular margin usage. After some testings, only SpotCrude has sustained performance for the past decade. The equity change is due to a new deposit, the numbers could be larger than expected due to this change.
On average, I am aiming about 2 - 4% profit per month, but like I mentioned below, the drawdown will always be there. However, so far no tickers have truly reached 0 from my backtesting, they are all winning on average just less than SpotCrude.
On average, I am aiming about 2 - 4% profit per month, but like I mentioned below, the drawdown will always be there. However, so far no tickers have truly reached 0 from my backtesting, they are all winning on average just less than SpotCrude.
Yiheng Quan
The previous approach was too NAIVE, but I have figured out a potential improvement. A new version of F.U.N.N.Y is on the way with better margin usage as well. The win rate will still lock around 40% - 44%, and so far, SpotCrude is the only ticker with real edge. However, other ones are still valid, overall, it will still reach my goal at the end. Further optimisation will be applied per ticker, for example, we may block certain tickers on certain days, and flip only under the right condition. Drawdowns will still exist and be there all the time, this isn't what the current system can address sadly.
Yiheng Quan
F U N N Yシグナルについてのニュース
Further adjustments are required, simply having an inverse signal was a bad idea. I will pause the system next week to work on my new approach
Yiheng Quan
With the signal flipping, the profit factor could reach above 2, win rate improves away above 50% and sharpe ratio can increase from 0.06 -> 0.18. However, this doesn't indicate future performance but for a reference purpose only. With a larger account like 5000$, the sharpe could improve significantly because the system is opening larger lots for certain tickers, but it is less a priority for now. Better profit factor, win rate and less drawdown is the main focus.
Yiheng Quan
F U N N Yシグナルについてのニュース
Wednesday, Thursday and Friday will now produce a counter signal. This aims to reduce drawdown while having more profit
Yiheng Quan
Wednesday and Friday are having poor performance, I have flipped the signal on these days starting from tomorrow, it will be using this logic. This could improve the performance if the past results are stable enough.
Yiheng Quan
Preparing for an enhanced FUNNY Signal, this will take a while. It is also working towards a better reinforcement learning agent.
Yiheng Quan
Very large drawdown, it was almost 5% even without the 2x multipler, but for now, I will focus more on the latest version of the model, and let the signal trade on itself.
Yiheng Quan
Another 10% today. I have increased the internal multiplier, which means a heavier deposit load and that 1–2% drawdowns will become more common. From today, FUNNY Signal will have roughly 2× larger drawdowns. However, as a side effect, profits will also be doubled.
Yiheng Quan
Great AUDUSD short on Good Friday, but I also saw some problem with the spread exploded, will make my training stricter on SL.
Yiheng Quan
Another 3% profit while I was sleeping, can’t imagine I’ve recovered my 12% loss and even managed to withdraw some cash, all within this month, almost 4% just this week.
Yiheng Quan
Promising results with reinforcement learning, but there are cases the agent cannot handle. Per step is about one hour, but this profit isn't realistic. Lots of work to do still, but will take a short break.
Yiheng Quan
Yesterday, I witnessed a legendary short trade that recovered my entire losses in just three hours while it was raining here. This happened exactly when I needed the courage to continue this path. This AI model is still far from its true potential, so I will be working on reinforcement learning to further finetune my existing models.
Yiheng Quan
パブリッシュされたMetaTrader 5シグナル
F.U.N.N.Y stands for Fully Unstable Neural Network for Yield. This is a highly experimental fully automated intraday trading system across multiple tickers. The latest version has been deployed, it is further finetuned with algorithm to focus on sortino ratio & recovery factor