16 years of high annual returns with, NO yearly drawdowns, NYSE stocks, using canned technical indicators, buy at the open with stops(implied they are tight stops). I would heed the suggestion to trade just a few shares risking very little. I have some fun(useless) software that can combine those indicators, those stocks, for various hold times with various stops etc.... AND probabaly couldnt produce those results after millions of permutatioms.
@eurusdzn you summed it up pretty well My results are way too good to be true. I'm still risking a little bit of money on it, nothing I can't live without. My thinking process has to be way off somehow. My results are likely to be flawed based on things like survivorship-bias (using stock that existed 16 years ago) and other things I don't know about. I am and have been worried about look-ahead bias as well. But I can't say I have any as far as I can tell. There must be all kinds of stuff skewing the results. I just have to figure it out. I found this one problem so far, I never sell stock on the first trade day (the day I buy) when backtesting. With the live trading I have exited stock on the first trade day. It's a limitation of the backtesting software, the way it is designed. You can't buy and sell the same stock on the first day. I have worked on that today and added a stoplimit on the first buy day at the same amount of the trailing stop. The stoplimit stays around for the entire duration I keep the stock if it does not get triggered, but as it's the same as the trailing stop and the trailing stop can only go up that does not seem to do any harm. I haven't backtested this yet. I doubt this will affect my results much. So yeah I will definitely agree my results are way too good to be true, so this is an interesting experiment and I will see where it goes.
My backtesting code broke, with the recent changes to accomodate selling on trade day. I managed to put in much better code today and I think I am able to run it properly now. I am doing some testing now, results look worse. Results look negative actually. It looks like the sell offs on the first trade day are very bad for the strategy. I added some code to have the stop be larger on day #1 (a multiplier of the normal stop). That means for actual trading I will need to add a stop just for day #1 and a trailing stop on the second day. I will do testing and see where it ends up.
Ok backtesting suggests I should never exit trade on day #1 (with my particular strategy). I will continue to test a little further but that is what I am going with at the moment. So what I will probably do is put in a stop on day #1 that never gets hit (or very rarely) over the last X years of backtesting. Because I do feel more comfortable having a stop than not having any. The stop might have to be around 12-15% or so. The Time in Force for that stop will be DAY. Then I need to put in my regular stop after trading on Day 1 before trading Day 2 with GTC (Good till Cancelled).
I hope nobody feels I am spamming them with this thread. I did create it in "Journals", not sure why it was moved away from there. Well.. I have been thinking with only being in the market at 4K and there going to be several hundred transactions per year (300?) and the transaction cost probably being at least 1USD/transaction I will have to weigh the results excluding transaction costs (or at least divide transaction costs by 10 the same as I did with starting capital). Just something I am going to have to keep in mind. Another thing I noticed is my transactions are less than 1USD average according to the backtesting software, I thought my broker transactions were at least 1USD/transactions so I may have to check into that too. I have a function to calculate transaction costs I guess it may be off.
One stock sold, I put in a new order with a trailing stop at -20%. I will adjust it tomorrow. Come to think about it I should try to find the exact stop where the trailing stop never gets hit. -20% is too high it is probably around -15%. Oh and the transaction costs seem to be working properly. Could not reproduce average transaction cost under 1USD. I have this problem with my backtesting software though where it is now out of sync with my live trades. Not sure how to get it in sync unless I figure out how to enter specific trades on certain days. I think I can figure that out.
One stock sold again, I will put in a new order today. I updated the Day #1 stop on yesterday's buy to the lower %. My total portfolio value is -5USD off from the start. Which means I am up if I discounted commissions. Buy and keep would have performed better. I guess I am impatient by nature I want to see insta-results. It's hard to accept I need to run this for 6-12months to get to a real level of confidence. Sigh.. long wait. I did have one interesting thought and that would be to modify the strategy to use intraday data. Maybe it would work or maybe not. I haven't looked at intraday data. My backtesting tool is written around daily data. It's suitable for intraday data too, *BUT* I have made some customizations and changes and never taken intraday data into account. I would have to look at where to purchase quality intraday data as well. I am using free Yahoo EOD data now. It seems accurate enough. There are practical concerns with trading intra day such as timezone difference, but maybe trading intraday would accelerate results? I think they would. I would know I had a winner or loser strategy in weeks not months. PS this is long term no immediate plans to make this intraday
This is interesting.. looks like the "sell stock on Day #1 (Buy Day)" or NOT "sell stock on Day #1" is significant in back testing in particular for 2008. The effect is negligible in other years. In 2008 results (under certain conditions that I use for back testing) for the year are +10% when NOT selling on Buy Day and as low as 0% (flat) with a 20% stop. That is counter intuitive to what I might have guessed, but if I think about it longer it makes sense too (maybe?). So now I have to ask myself can I live with no stops on Day #1. Probably I could, but only if I purchased one or two new stocks per day maximum (out of 10). I haven't looked closely enough at 2008, but I have a feeling it currently is selling and buying multiple stocks on the same day. Sigh.. more back testing to figure out the optimum or just put in a back stop at 15 or 20% and accept my results will be much flatter (or negative) if another 2008 comes around.
Account is in the plus now even with commissions being rather a big part of the 4K capital. One stock got sold, I will buy another one today. Looks like I need to put Buy Day stop quite high for best results. I chose a percentage I will go with and I think I can live with it (it's above 20% but below 30%). I might not do updates on portfolio status every day unless something big changes. I expect it will hover around break even and slowly creep up, unless market volatility sky rockets then I expect big changes (up or down we will see but hopefully up on average). One thing I am working on is this.. it is difficult to keep portfolio in sync with my backtesting software. It is already out of sync because some stock got sold off that my backtesting software would have kept. Another thing that would get it out of sync is that I use X stocks but actually I sometimes put in X+1 to go slightly over 4K. I only do that if I see the cash in the account is ... Hey wait stop the presses.... I notice I've only been investing 2K instead of 4K. I put in 42K and I intended to go in 2K initially for testing, but somehow made that figure 4K in my mind. What to do.. I think I will make it 4K now that means I'll be able to start putting in more stock from today (20 or 21 instead of 10 or 11). By the way I feel pretty stupid about this. I was just thinking this morning I have trouble with mathematics I never managed to do integrals and what not. That fits into another thing that happened today and yesterday, I started looking into neural nets (after reading on this website). I looked and read a little before but this time I followed some of the steps to build a small neural net and it all became kind of real for me. I don't understand a lot of the complex mathematics behind it, but I'm good at troubleshooting and making stuff work and making changes to it to make it run better and do what I want. So that will definitely be a next project, fit some prices and/or indicators into input and setup a modest neural network and play with it a little and then when I have the basic concept going use some insight to make it do something nobody has made it do yet. Haha we will see, but yeah that would be the next project. Actually not understanding a lot about something can sometimes lead to new insight. The first thing I need to work on today though is I want to add a "starting position" into my backtesting software so I can let it run with where I am at with the live account which is now out of sync. Not sure whether that will work or quite how, but we will see.