Trading idea- Based on Graham (TSX)
I've built this portfolio to trade Canadian equity on TFSA. This trading model does the screening of companies according to Graham's principle, as per buy rules below. The model then chooses the best 10 companies based on a ranking also described below.
The model
This is a low-turnover model, rebalance is at every 4 weeks only (current holdings are evaluated if they need to be sold or held at every 4 weeks). This model uses market timing, and that signal is checked weekly. Also, when the model is in cash (not fully invested), it checks weekly if there's any company to be purchased. I will be posting the transactions here so that one can use it for further due-dilligence. I'm trading this model as well.
This model is based mostly on fundamentals, and it has been revised to contain a market timing rules, besides momentum criteria for the ranking. According to backtests, the model avoid big drawdowns, however only time will show effective this market timing rule can avoid future drawdowns caused by crashes. The model has been live for a while now, and I've been simply following it - but feel free to do your due-dilligence if required.
This model has been revised on July 4, 2017 to include a rule filtering by stocks with higher volume, to allow the model to scale better and provide decent liquidity.
The rules
The universe for this model are the stocks in TSX only - so no Venture Exchange companies.
Buy rules:
- Price > $5 (no penny stocks);
- Current ratio is at least 1.5;
- Long term debt is less than 110% of working capital;
- Last 4 quarters of EPS above breakeven;
- Last 5 years of EPS above breakeven;
- Annual EPS grew over past year and past 5 years;
- Company has paid dividends within past year.
- Rank >= 80
- Average volume from last 60 days > 100,000
The top 10 companies of the ranking are selected.
Sell rule:
- Ranking < 80
The ranking is built by having equal weight distribution (25%) to the following criteria: Value, Growth, Quality and Momentum.
Value: 65% based on income stream (Current Fiscal Year Projected P/E Ratio, Projected Price/Earnings to Long Term Growth Rate and Market Cap to Adjusted earnings) and 35% based on others (Enterprise Value to Sales, Enterprise Value to Enterprise Asset and a conditional PB that will be applied only if 5-year ROE is above the industry average). These parameters consists the new ranking, updated for November 2015 rebalance, and described here.
Growth: 75% based on EPS growth (change per quarter, TTM and 5 years) and acceleration (formula for recent and long term acceleration ratio) and 25% based on sales growth and acceleration (same interval as EPS growth and acceleration).
Quality: 25% based on operating margin % (TTM and 5-year average), 25% based on asset turnover, 25% based on ROI and ROE (TTM and 5-year average) and 25% based on Finances (current ratio, interest coverage, total debt to capital ratio).
Momentum: 65% based on price changes (ratio between today and different past periods) and 35% based on technical indicators (up down ratio for different periods).
Market Timing rules:
The model relies on estimated earnings for the SP500 companies to enter or exit the market. History shows that when earnings from the biggest 500 US companies decline, the market declines globally (and vice-versa), so it's more reliable than comparing with earnings from Canadian index, which is too heavy on financials and energy companies, and which the model rarely trades on. Therefore, the model attempt to exit the market when price is estimated to decline due to earnings - not to avoid declines on news, like it happened in 2011 or on a smaller scale, in 2015. History has shown these are short-lived anyway.
First, a daily graph of blended estimated earnings are plotted. This is calculated using a blend of the Current Year and Next Year estimates for the S&P500 stocks. The weight given to the Current Year and Next Year depends on which quarter was the most recent. For each SP500 stocks the following Blended_EPS variable is calculated:
if we're on Q4 then it uses the Current Fiscal Year estimated EPS mean;
if we're on Q1 then it uses 75% of the Current Fiscal Year estimated EPS mean and 25% for the Next Fiscal Year estimated EPS mean;
if we're on Q2 then it uses 50% of the Current Fiscal Year estimated EPS mean and 50% for the Next Fiscal Year estimated EPS mean;
if we're Q3 then it uses 25% of the Current Fiscal Year estimated EPS mean and 75% for the Next Fiscal Year estimated EPS mean.
The quarters refer to calendar year.
This SP500 Blended_EPS variable above is then computed as (for each stock as "i" from SP500):
num = Sum { Blended_EPS (i) * Shares (i) }
den = Sum { MktCap (i) }
Plotted value: ( S&P500 Close * (num/dem) )
By plotting this daily, a graph is formed. Then, a known strength technical oscillator is used to enter or exit the market when the graph hits a certain threshold value - going to cash when strength confirms a weak signal (which reflects a lower blended earnings, which should move stock prices lower), or going back to stocks when the strength oscillator confirms a stronger signal (which reflects a higher blended earnings, which should move stock prices higher).
How to follow
Please do your due-diligence before following this (or any) model. Ensure that you understand what the rules are about and if it fits your risk tolerance. Even though this model uses market timing, there will be times that the model will have paper and realized losses. There might be times that the model under perform the index - it doesn't mean that the strategy is broken. Since the model is based on fundamentals, long term results should be superior, but only time will tell. Also, out-of-sample record of this model is very small - only time will demonstrate the robustness of this model, not only from keeping the same curve in out-of-sample, but also how it will endure the next crash.
To mimic the performance of this model, one would equally buy the current holdings, hence the suggestion to start with at least $10k ($1k each). I personally prefer to start when the model was rebalanced, even though if stocks have huge gains. If the model doesn't have a sell signal, it's very possible to continue appreciating in value. Since one might buy a stock during rebalance that is being held for a long time, one might be subject to losses if that stock declines when it's sold (while the ones that bought in the original buy signal might not experience that loss, because they bought way earlier). However, the longer one stick with the model, the more one will allow its effect to take place, so I'd wait at least 1 year to abandon a model (that's my personal approach). Also, if a model had losses in that year, I'd compare it with the benchmark to see if it did better than the market, even if both had losses.
There will be paper and realized losses, and that's by design, because the model does not (and cannot) use subjective decisions to buy or sell, it's all automated, and therefore, like any mechanical model, not perfect. However, I expect it to have double-digits annualized return, hence I'm invested on it.
Update on modified version of this model
As per discussion on this post and on this post, the same model without the criteria for dividends would place growth companies in the mix, and have a higher adjusted risk return as well as total performance, although the model would be more volatile. This model is free for anyone to copy and if anyone is interested in modifying the rules to eliminate the dividend screening, I can help guide through these steps. I no longer have a live model slot available to post this variation model, so I'll keep the original model live (and will continue to post the transactions here), while maintaining the simulated model public to offer the option to anyone to modify this model as they wish, including removing the dividend screening in portfolio123. My remaining slots are being used to my other trading models.
A second update took place on July 4, 2017, to scan for stocks with higher liquidity. Backtests shows better results (higher annualized returns, same drawdown and higher Sharpe when compared to the original model which didn't take liquidity into account).
Original model:
Next rebalance date
August 21, 2017
Last signals (August 14, 2017):
Sell: XTC, MAL
Buy: CWX, SJ
The model holds up to 10 stocks and it will be 100% invested (0% in cash) after today's rebalance.
Current holdings (August 13, 2017):
The model is currently 100% invested (0% cash).
Information since inception
Summary Info: (Updated Live)
https://www.portfolio123.com/embedded_p ... chartW=450
Performance info: (August 13, 2017)
Stats info:(August 13, 2017)
Risk measurement info: (August 13, 2017)
Allocation info: (August 13, 2017)
Ranking info: (August 13, 2017)
FAQ
1. What does "rebalance" mean?
Rebalance is the execution of the screening rules to determine if a current holding should be sold or kept. This is done at every 4 weeks. Market timing signal is verified weekly. If the model is not fully invested, then it checks weekly if there are any candidates that meet all criteria. It will contain new sell / buy signals if there's anything to be sold, or it specifically indicate that no action is required (nothing to sell, therefore, nothing to buy).
2. Is it too late to follow it? When should I start?
I would personally wait until the next rebalance day to start, and buy its holdings spread equally. Be aware of your comission cost, to ensure that you have enough to trade to make sense.
3. What is the definition of "EPS above breakeven" used by the screener?
It means positive EPS for all of the most recent 4 quarters (each one of the last quarters must have had positive EPS). For that, I use EPS Excluding Extraordinary Items, which is earnings per share including all expense with the exception of those deemed extraordinary. This is calculated with net income, including all expenses but excluding extraordinary items, and using fully diluted shares (the most common source of extraordinary items in recent accounting is discontinued operations). The data points come from SEC filings or earnings announcements of a company, and it's calculated by Compustat, which is used by portfolio123 platform.
Comments
EDIT (2/27/2015): clarification: rebalance is at every 4 weeks. So there will be 13 signals in a year, not 12 (I had marked earlier that rebalance was monthly).
EDIT (3/21/2015): Added links for performance, stats and risk measurement info. This will be updated every 4 weeks, before the next rebalance.
EDIT (3/22/2015): Added link for allocation information.
EDIT (3/31/3015): Added info for recent transactions and current holdings.
EDIT (11/2/2015): Updated ranking system details on the first OP.
EDIT (4/4/2016): Updated description regarding market timing rules.
EDIT (4/13/2016): Updated FAQ.
EDIT (4/24/2016): Details on additional posting for modified model (no dividend requirement)
EDIT (1/22/2017): Updates for the modified model.
EDIT (1/29/2017): Clarification on rebalance frequency.
EDIT (7/4/2017): Updated model with liquidity rule.
Rod
The model
This is a low-turnover model, rebalance is at every 4 weeks only (current holdings are evaluated if they need to be sold or held at every 4 weeks). This model uses market timing, and that signal is checked weekly. Also, when the model is in cash (not fully invested), it checks weekly if there's any company to be purchased. I will be posting the transactions here so that one can use it for further due-dilligence. I'm trading this model as well.
This model is based mostly on fundamentals, and it has been revised to contain a market timing rules, besides momentum criteria for the ranking. According to backtests, the model avoid big drawdowns, however only time will show effective this market timing rule can avoid future drawdowns caused by crashes. The model has been live for a while now, and I've been simply following it - but feel free to do your due-dilligence if required.
This model has been revised on July 4, 2017 to include a rule filtering by stocks with higher volume, to allow the model to scale better and provide decent liquidity.
The rules
The universe for this model are the stocks in TSX only - so no Venture Exchange companies.
Buy rules:
- Price > $5 (no penny stocks);
- Current ratio is at least 1.5;
- Long term debt is less than 110% of working capital;
- Last 4 quarters of EPS above breakeven;
- Last 5 years of EPS above breakeven;
- Annual EPS grew over past year and past 5 years;
- Company has paid dividends within past year.
- Rank >= 80
- Average volume from last 60 days > 100,000
The top 10 companies of the ranking are selected.
Sell rule:
- Ranking < 80
The ranking is built by having equal weight distribution (25%) to the following criteria: Value, Growth, Quality and Momentum.
Value: 65% based on income stream (Current Fiscal Year Projected P/E Ratio, Projected Price/Earnings to Long Term Growth Rate and Market Cap to Adjusted earnings) and 35% based on others (Enterprise Value to Sales, Enterprise Value to Enterprise Asset and a conditional PB that will be applied only if 5-year ROE is above the industry average). These parameters consists the new ranking, updated for November 2015 rebalance, and described here.
Growth: 75% based on EPS growth (change per quarter, TTM and 5 years) and acceleration (formula for recent and long term acceleration ratio) and 25% based on sales growth and acceleration (same interval as EPS growth and acceleration).
Quality: 25% based on operating margin % (TTM and 5-year average), 25% based on asset turnover, 25% based on ROI and ROE (TTM and 5-year average) and 25% based on Finances (current ratio, interest coverage, total debt to capital ratio).
Momentum: 65% based on price changes (ratio between today and different past periods) and 35% based on technical indicators (up down ratio for different periods).
Market Timing rules:
The model relies on estimated earnings for the SP500 companies to enter or exit the market. History shows that when earnings from the biggest 500 US companies decline, the market declines globally (and vice-versa), so it's more reliable than comparing with earnings from Canadian index, which is too heavy on financials and energy companies, and which the model rarely trades on. Therefore, the model attempt to exit the market when price is estimated to decline due to earnings - not to avoid declines on news, like it happened in 2011 or on a smaller scale, in 2015. History has shown these are short-lived anyway.
First, a daily graph of blended estimated earnings are plotted. This is calculated using a blend of the Current Year and Next Year estimates for the S&P500 stocks. The weight given to the Current Year and Next Year depends on which quarter was the most recent. For each SP500 stocks the following Blended_EPS variable is calculated:
if we're on Q4 then it uses the Current Fiscal Year estimated EPS mean;
if we're on Q1 then it uses 75% of the Current Fiscal Year estimated EPS mean and 25% for the Next Fiscal Year estimated EPS mean;
if we're on Q2 then it uses 50% of the Current Fiscal Year estimated EPS mean and 50% for the Next Fiscal Year estimated EPS mean;
if we're Q3 then it uses 25% of the Current Fiscal Year estimated EPS mean and 75% for the Next Fiscal Year estimated EPS mean.
The quarters refer to calendar year.
This SP500 Blended_EPS variable above is then computed as (for each stock as "i" from SP500):
num = Sum { Blended_EPS (i) * Shares (i) }
den = Sum { MktCap (i) }
Plotted value: ( S&P500 Close * (num/dem) )
By plotting this daily, a graph is formed. Then, a known strength technical oscillator is used to enter or exit the market when the graph hits a certain threshold value - going to cash when strength confirms a weak signal (which reflects a lower blended earnings, which should move stock prices lower), or going back to stocks when the strength oscillator confirms a stronger signal (which reflects a higher blended earnings, which should move stock prices higher).
How to follow
Please do your due-diligence before following this (or any) model. Ensure that you understand what the rules are about and if it fits your risk tolerance. Even though this model uses market timing, there will be times that the model will have paper and realized losses. There might be times that the model under perform the index - it doesn't mean that the strategy is broken. Since the model is based on fundamentals, long term results should be superior, but only time will tell. Also, out-of-sample record of this model is very small - only time will demonstrate the robustness of this model, not only from keeping the same curve in out-of-sample, but also how it will endure the next crash.
To mimic the performance of this model, one would equally buy the current holdings, hence the suggestion to start with at least $10k ($1k each). I personally prefer to start when the model was rebalanced, even though if stocks have huge gains. If the model doesn't have a sell signal, it's very possible to continue appreciating in value. Since one might buy a stock during rebalance that is being held for a long time, one might be subject to losses if that stock declines when it's sold (while the ones that bought in the original buy signal might not experience that loss, because they bought way earlier). However, the longer one stick with the model, the more one will allow its effect to take place, so I'd wait at least 1 year to abandon a model (that's my personal approach). Also, if a model had losses in that year, I'd compare it with the benchmark to see if it did better than the market, even if both had losses.
There will be paper and realized losses, and that's by design, because the model does not (and cannot) use subjective decisions to buy or sell, it's all automated, and therefore, like any mechanical model, not perfect. However, I expect it to have double-digits annualized return, hence I'm invested on it.
Update on modified version of this model
As per discussion on this post and on this post, the same model without the criteria for dividends would place growth companies in the mix, and have a higher adjusted risk return as well as total performance, although the model would be more volatile. This model is free for anyone to copy and if anyone is interested in modifying the rules to eliminate the dividend screening, I can help guide through these steps. I no longer have a live model slot available to post this variation model, so I'll keep the original model live (and will continue to post the transactions here), while maintaining the simulated model public to offer the option to anyone to modify this model as they wish, including removing the dividend screening in portfolio123. My remaining slots are being used to my other trading models.
A second update took place on July 4, 2017, to scan for stocks with higher liquidity. Backtests shows better results (higher annualized returns, same drawdown and higher Sharpe when compared to the original model which didn't take liquidity into account).
Original model:
Next rebalance date
August 21, 2017
Last signals (August 14, 2017):
Sell: XTC, MAL
Buy: CWX, SJ
The model holds up to 10 stocks and it will be 100% invested (0% in cash) after today's rebalance.
Current holdings (August 13, 2017):
The model is currently 100% invested (0% cash).
Information since inception
Summary Info: (Updated Live)
https://www.portfolio123.com/embedded_p ... chartW=450
Performance info: (August 13, 2017)
Stats info:(August 13, 2017)
Risk measurement info: (August 13, 2017)
Allocation info: (August 13, 2017)
Ranking info: (August 13, 2017)
FAQ
1. What does "rebalance" mean?
Rebalance is the execution of the screening rules to determine if a current holding should be sold or kept. This is done at every 4 weeks. Market timing signal is verified weekly. If the model is not fully invested, then it checks weekly if there are any candidates that meet all criteria. It will contain new sell / buy signals if there's anything to be sold, or it specifically indicate that no action is required (nothing to sell, therefore, nothing to buy).
2. Is it too late to follow it? When should I start?
I would personally wait until the next rebalance day to start, and buy its holdings spread equally. Be aware of your comission cost, to ensure that you have enough to trade to make sense.
3. What is the definition of "EPS above breakeven" used by the screener?
It means positive EPS for all of the most recent 4 quarters (each one of the last quarters must have had positive EPS). For that, I use EPS Excluding Extraordinary Items, which is earnings per share including all expense with the exception of those deemed extraordinary. This is calculated with net income, including all expenses but excluding extraordinary items, and using fully diluted shares (the most common source of extraordinary items in recent accounting is discontinued operations). The data points come from SEC filings or earnings announcements of a company, and it's calculated by Compustat, which is used by portfolio123 platform.
Comments
EDIT (2/27/2015): clarification: rebalance is at every 4 weeks. So there will be 13 signals in a year, not 12 (I had marked earlier that rebalance was monthly).
EDIT (3/21/2015): Added links for performance, stats and risk measurement info. This will be updated every 4 weeks, before the next rebalance.
EDIT (3/22/2015): Added link for allocation information.
EDIT (3/31/3015): Added info for recent transactions and current holdings.
EDIT (11/2/2015): Updated ranking system details on the first OP.
EDIT (4/4/2016): Updated description regarding market timing rules.
EDIT (4/13/2016): Updated FAQ.
EDIT (4/24/2016): Details on additional posting for modified model (no dividend requirement)
EDIT (1/22/2017): Updates for the modified model.
EDIT (1/29/2017): Clarification on rebalance frequency.
EDIT (7/4/2017): Updated model with liquidity rule.
Rod
Last edited by rodbarc on Feb 22nd, 2015 10:44 pm, edited 64 times in total.
