Formal Rules for Stops

I have decided on formal rules for setting daily stop losses. It is based on the pivot-point method. The pivot point is the average of the high, low, and close for the previous day. When short the stop loss is set at the second resistance level - the pivot point plus the previous day's high-low range - and when long it is set at the second support level - the pivot point minus the previous day's high-low range. If this results in a stop that is less than 1% from the opening price, I instead set a 1% stop. These stops increase the Sharpe ratio of the model though they slightly decrease returns. The chart below shows the last month of daily pivot points and first and second resistance levels:

The model got stopped out on 26 April when short - losing 1.42% that day. The market closed up 2.08%. So that saved 0.66% of losses The model also got stopped out on 3 May when long losing 1%. The market closed only down 0.02%. So that increased loss by 0.98%. That shows you why this reduces returns...

These numbers don't quite match what you can see on the chart as the chart shows the 24/5 futures market and the model is based on the NASDAQ 100 index. I am thinking of switching the model to use actual futures prices. Will need to pay for the data, I think.

Trying to Learn the Lesson about Narrow Stops Again

Yesterday the model said to go short NQ (NASDAQ) and long ES (S&P500). I started off the day, doing exactly that, though I entered the trade badly and ended up down on the NQ part of the trade relative to the ES part of the trade. Then, I closed the ES long for a small profit and based on "pivot points", I set a stop loss at 7010 - 50 points above my entry point. As you can see from the chart, the market briefly went through the stop but then turned and ended the day near where it started. So, I lost a lot more money than the model did. If I had set the stop at 1% (7030) or kept the hedge without stops, I would have ended the day with only a small loss. Really, it was fear of missing out on making a profit on the ES trade that screwed me up.

I also did a long CL (crude oil) trade yesterday and was stopped out too. Then the market also turned around and actually ended the day higher. The lesson I am learning from that one is to stick to the stock market for the moment so I can set wider stops and not worry about the potential loss from multiple positions going wrong.

The model is again short NQ and long ES for today. I was going to sit out for today, but writing this post has inspired me to put on that hedged trade. For the record, I sold NQ at 6969.75 and bought ES at 2729.50.

Weekly Update

Another successful week. Now up USD 8k for the month and USD 6k since the trading experiment started. On Friday I was long NQ when I should have been short. I only lost USD 150 luckily (though double that relative to what I would have got if I had done the correct trade). This was because of an error in a link in a spreadsheet. That link is now fixed. The model says short NQ (NASDAQ), long ES (S&P 500), long CL (Crude) for Monday. I think the short NQ is only a short term correction in NQ and probably it will switch back to long by Tuesday. It is a bit of an unusual feeling to see myself keep winning trades. I said to Moominmama that it felt like I was cheating or something. She said: "Please don't feel like that, please make lots of money :)". So far, this month I just have had to stay long, which isn't so easy for me as I tend to be bearish. And apart from Monday it looks like that staying long will continue to be the challenge for a little while till the model actually shifts to the short side. So, for the meantime we are still in phase 2 of this experiment, which is to see if I can stick to what the model says to do. Only, when we've been through both a long and a short phase successfully, will we be able to say that I think.

These 13F Tracking ETF's Have Horrible Performance

13F is a form lodged quarterly by US based investment funds. A 13F following strategy takes the stock picks from top hedge funds as revealed by their 13F forms. Two ETF's that follow this strategy are ALFA and GURU. But both have horrible performance with negative alpha of of -5% and -7%, which is rather ironic. Does this strategy no longer work?

Cracking Horse-Racing, the Lottery, and the Stockmarket?

Articles about Bill Benter who "cracked" gambling on horse-racing by using a model to predict which horses would win and Eddie Tipton who cracked the state lottery, illegally. I'm testing whether I've cracked the stock market :) So far, so good this month, but it is early days.  

P.S.
More on quant betting on horse-racing. Model remains long stocks (NDX and SPX) and switches to long oil for Monday. Yes, I added a model for predicting oil, so far I only did very quick trades in oil.

P.P.S.
More on Zeljko Ranogajec.

Fear of Missing Out versus Loss Aversion

The key to sleeping better in Australia while trading in the US markets seems paradoxically to be using wider stop losses rather than tighter stop losses. With a tighter stop, I am concerned that the market will hit the stop and then bounce back up strongly, which is what would have happened last night except I stayed up and adjusted the stop. This is the fear of missing out - crystallizing a loss and then missing the upside. I need to be more accepting of the possibility of large losses to allow the possibility of gains. I actually seem to have less aversion to losses if they aren't tied to then missing out on gains. FOMO seems to beat loss aversion. This is because my trading model has a high win rate. Traders with techniques that have a small edge or no edge have to make sure that wins are bigger than losses - letting winners run and cutting losses. They need the asymmetry to make money. I don't.

April 2018 Report

A very active month financially. The Australian stock market rebounded quite strongly and now looks pretty bullish to me. I also started trading futures again, which so far had the opposite effect on the results for the month :)

The Australian Dollar fell from USD 0.7680 to USD 0.7540. The MSCI World Index rose 1.08%, and the S&P 500 0.38%. The ASX 200 rose 3.92%. All these are total returns including dividends. We gained 2.86% in Australian Dollar terms and 0.98% in US Dollar terms. So, we underperformed the Australian market and to a small degree the international markets but outperformed the U.S. market.

The best performing investment in dollar terms was CFS Geared Share Fund up AUD17k. The worst performer in dollar terms was IPE, down AUD3k. My holding is now quite large (more than 1% of the value of the company - it's a very low value company) and the price is quite erratic. The best performing asset class was large cap Australian stocks, which gained 2.84%. The worst performing asset class was private equity, losing 2.04%, the only asset class to lose money this month.

A new item that I am reporting from this month is trading income. This includes trading in futures and options etc and interest on cash dedicated to trading. It doesn't include any trading done on fundamental grounds. This month I lost money - USD1,987 - which isn't surprising as I was experimenting with different models and approaches and learning to trade more confidently. I pretty much reversed that on the first day of this month, but anything could happen. Less than 3% of net worth is dedicated to trading at this point, which mainly means a deposit of Australian and US dollars used as margin for derivatives. The plan for this month is to consistently trade one futures contract according to the trades that the model provides, while learning about entering trades more optimally and setting stops or using options as hedges (much wider hedges than I was using last month).

We made a bit more progress towards the new long-run asset allocation:

Total leverage includes borrowing inside leveraged (geared) mutual (managed) funds. The allocation is according to total assets including the true exposure in leveraged mutual funds. I have reduced the allocation to cash, because assuming I will be trading, there will always be plenty of cash in the trading account plus the ability to borrow, though the latter can be reduced in a financial crisis. Commodities now includes managed futures, trading, and gold.

The "improvement" in allocation, came partly due to market movements and partly due to investment activity. We invest AUD 2000 monthly in a set of managed funds, and there are also retirement contributions. Then there are distributions from funds and dividends. During the month, I also:

• Invested in a venture capital fund.
• Bought more IPE (private equity) at below net asset value.
• Sold out of Leucadia National (LUK) and bought more 3i (III.L, private equity) and China  Fund (CHN).
• Bought more units in the Winton Global Alpha fund (managed futures - in the commodities category).
• Transferred cash into my trading account and did a lot of trading of futures and options while developing my trading model.

As a result the allocation to private equity and commodities increased quite a bit.

New Model Rule

The model was still short yesterday based on the forward forecast.  But Facebook's earnings release pushed the market up and I lost money. So, now I will check the model in real time as well before putting on trades. Also, today's decision based purely on forecasts would have been to stay short like yesterday's. But yesterday the actual observed indicator signalled a buy. So, now we add a rule that if the actual observed signal yesterday was a buy that over-rides a forecast....

Now Amazon released their earnings an the market is up in after hours again.... I put on my first full size futures order with a 1% stop buying the E_Mini S&P. Now this is the real test of the model...

Sophisticated Investor

I got an e-mail about an Australian venture capital fund and decided to follow it up. The information the fund sent me looked very interesting, but it is limited to wholesale and sophisticated investors. In order to be classified as a wholesale investor you must have individually (not with your spouse) AUD 2.5 million in net assets or AUD 250k in gross income. I don't qualify individually on this basis, though we jointly would qualify on the second criterion and in the near future I will qualify on the first criterion. So, I told the fund salesperson that and they sent me a questionaire to see if I qualify as a sophisticated investor who understands the risks involved. I just sent the form back. If they qualify me I will invest in the fund and disclose more information here. Overall, I plan to invest 5% in private equity and it makes sense to allocate half of that to venture capital and half to buyout etc. IPE and OCP cover the later stage private equity in the portfolio 2.5% roughly equals the fund's minimum investment requirement, so that is what I will invest, if approved. Interestingly, early stage venture capital investments are tax free in Australia. That also means, of course, that you can't claim losses against your income tax.

In other news, I redesigned a trading algorithm from the bottom up on 2018 data, using the same forecasting model. It has a bit lower return and larger drawdowns, but all the rules make theoretical sense and it sticks to the model predictions rather than reversing direction if stopped out. In fact, it only uses a stop when initiating a new direction - this is to guard against the new signal being noise - the stop is removed after the direction is confirmed. After that I would just use hedges. Next, I need to backtest it for 2017 and 2007. I think 2007 is analogous to 2018, while 2017 is very different - a constantly uptrending market.

The model is currently short, but I am not trading it without backtesting and also there is higher risk entering a move already underway, as the model is unlikely to time the exact optimal turning point to reverse direction.

P.S. 25 April
I backtested the model for the second half of 2017. Results are not as good as year to date in 2018 but they are much better than the model I was using at that time when the fake stops are removed from the model. The main issue is that my model tends to underperform the market in strongly trending markets as it keeps looking for opportunities to go short. We can compensate for this by trading 2/3 the model and 1/3 just long the index. This means that when we go long we use 3 times the position we use when we go shorter. This results in a more consistently rising equity curve. Increasing position size when going in the direction of the established trend definitely makes sense.

P.S. 27 April
They accepted me as a sophisticated investor.

Backtesting Failed

I backtested the model for 2017 and some periods in 2006-8 and it either makes no money (2017) or matches the market. So, this needs a fundamental rethink to get a more robust model. For the moment I will step away from live trading and observe what happens and when I have more time have a look again and developing a robust framework. I also saw that some of what I put in spreadsheets from 2006-08 is not realistic regarding how stops work, and so exaggerates the performance of the model. The current model was stopped out very very often in that period as the index was often both up and down more than 1% in a day. It could only work with hedges rather than hard stops.

Improved Model

The "Gold Model" was stopped out twice in a row on Thursday and Friday when the market was more than 1% down and it was long. So I now took another of my old models that uses a different approach entirely but only gives rare signals. Those signals over-ride the "Gold Model" signals. The combination enhances return. It would have been short Thursday and long Friday. It signals short for Monday. In simulation, it's average win is 1.3% a day and average loss 0.36% with a 62% probability of winning.  The Sharpe ratio is 0.52. This is only based on data since January 1st. Trading 1 NQ contract the biggest drawdown since January 1st is USD 3852.

I ended up USD 200 down on the month in trading at the end of Friday after being USD 700 up on Wednesday. Plan is to switch to trading NQ contracts with a stop loss next week. I am a bit wary of taking the signal from the new model for my first larger trades and so maybe will wait till Tuesday.

I made decision trees in Powerpoint for using the model in each of the 4 possible states where yesterday's trade was: long, short, long but stopped out, and short but stopped out. There is then no discretion over putting on trades. Here is one of the four decision trees, to give you an idea:

Collared Trading Has a Low Expected Value

I did a proper analysis of trading futures with an options collar. The win and loss probabilities are the same as trading with a stop loss. But the average win is reduced from 1.26% to 0.73% and the average loss from -0.65% to -0.53%. As a result the expected value of each day's trade goes down from 0.57% to 0.28%. And that's with perfect execution of the hedges and entry point so that the hedge is costless and the upside and the futures entry point is exactly in the middle of the interval between the hedges. Usually the hedge costs something, maybe 0.1% and the entry point isn't perfect. Together this probably reduces the expected value to a 0.1% gain or so. Some slippage in entry point on the futures plus stop tactic doesn't have such a big effect on the expected value. Maybe reducing it to 0.5%.

Given this, I have no choice but to bit the bullet and trade futures with a stop loss instead of a hedge and accept the relatively larger dollar value of losses when stopped out, as would have happened today trading NQ.

Why is Trading Stocks So Expensive in Australia?

Commonwealth Securities Charges 0.1% and Interactive Brokers 0.08% for Australian stock trades. That means that to trade AUD 170k of stock you would pay AUD 136 with IB. But to trade an S&P 500 futures contract costs USD 2.05 at IB. For U.S. stocks IB charge 0.5 cents per share so trading the same value of the SPY ETF costs USD 2.50.

These high prices aren't unique to Australia. IB charges around 0.1% to trade shares in most countries apart from the U.S. and Canada. On the other hand they charge AUD 1-6.5 per contract for Australian futures. So, maybe the question, should be why U.S. and Canadian shares are so cheap to trade.

Anyway, the high prices definitely discourages day-trading of Australian shares.

Long Futures Collar Trade

I put on my first collared futures trade. The idea was to sell a call 5 points above my long futures entry point and buy a put 5 points below. But my futures entry was at 2676.25 instead of 2675 and and the call was 1.5 points less than the put. So my potential upside is only $112.50 not counting fees and my potential downside is $387.50. As the model has a 71% win rate, the expected value is -$32.50 :( It's probably worse than that because the futures gapped up over the weekend reducing the potential upside. Oh well, the expected "tuition fee" is not so big. The screenshot shows the current state of play. Down.

P.S. 17 April
I "managed" the trade a bit and the futures were just below 2680 when the options expired. So I had a naked futures position, which I then sold at 2680.25. Overall, I made about USD 200 on the trade. I have now put another trade on. Long futures at 2681.5, sold a 2690 call for 7.25 and bought a 2675 put for 9.25. Maximum upside is USD 325 and maximum downside is USD 425 not counting commissions, which are small. Expected value is about USD 100. The spread between the two options today is 15 points, up from 10 points yesterday. The idea is to gradually widen the points spread as I am comfortable with it, eventually buying the put 25 points below the futures entry price, which is equivalent to a 1% stop. Yeah, the model is still long, the market is "overbought" and trending up according to the model.

ASX 200 and MSCI All World Total Returns

The Australia share price index - the ASX 200 - has not performed well since 2007. The current level is below its peak. However, when you add in both dividends and franking credits, it has almost doubled since the peak. Since 1996 it has returned twice as much as the MSCI in Australian Dollar terms, though since the crisis the two have had about the same gain, tripling from the low.

The Gold Model

I have now managed to reconstruct something similar to the old model I tried to trade a decade ago. It is a mixture of trend following when the markets are trending and predicting the direction to trade in when markets are more choppy. It follows a clear set of rules with no real discretion. Using those rules since January 1st this year would have returned 51% with a Sharpe ratio of 0.58. The model wins 71% of the time with an average daily win of 1.15%. The maximum loss is 1% as set by the stop. When I optimize a portfolio of the various methods I have come up with to maximize the Sharpe ratio of the portfolio the solution says to put 90% in this strategy and to actually short one of the other strategies! At the moment the model is long, which is good, as I have a long calls position still on from Friday. I think I will rename the new version of the old model the "gold model" :)

By the way, if you can borrow, maximizing the Sharpe ratio makes much more sense than maximizing return. You then get the smoothest time path of returns, which you can lever up if you want taking into account the size of likely drawdowns.

Friday Update

I finally exited Leucadia National after they announced they are converting into an investment bank and will sell some of their private equity assets and change their name to Jefferies. It has been one of my worst investments losing just under USD 4k (there have been some far worse ones though...). I got in too late just as the financial crisis was getting underway and the company never recovered. Previously, it had an excellent track record and was referred to as a mini Berkshire Hathaway. I bought some more shares in 3i instead, which has been a good investment.

I have been doing more work on improving my trading model and more on trying to trade it. The options I am willing to buy - i.e. the maximum loss is bearable - have too much time decay and so the market can go up and I end up not making money. I about broke even over the week due to this and various stupid things I continued to do. I stayed up last night trading the market, though that is not something I should do. The market was beginning to go up after initially falling and I decided for some reason that shorting put options on expiration day would be a good idea. I shorted a 2650 Friday 13th S&P put and bought a 2625 Friday 13th S&P put. This exposed me to a maximum USD 1300 loss. Of course, the market immediately turned around and started going down. As it reached 2650 I sold short a futures contract as a delta hedge. Then the market bounced and hit the stop I had subsequently put in place at 2650. And then it went back down again... In the end, I actually ended up about USD 20 on the trade :) But it was quite nerve-wracking. I don't know how ERN sells lots of put options all the time without trying to determine market direction or "buy reinsurance" to make it an options spread.

Next week I am thinking to experiment with futures contracts with options collars to limit both the upside and downside. Using options you can have a much tighter effective stop and not worry about the market coming down, hitting the stop, and then going up again. The downside is that the upside has to be limited or the cost of the put option (or call if short the futures) is too much. So, you sell a call (or a put) to defray the cost. I think for a reasonable net cost it's possible to have a little more upside than downside, though as the model does have an edge it's not strictly necessary to have more potential upside than downside. I am thinking of buying a put 10 points (E-Mini S&P) below the entry price into the futures contract and selling a call 20 points above. The maximum downside is then something like 13 points (USD 650) and the maximum upside 17 points (USD 850). Another downside is that if the market is flat, you lose 3 points (USD 150).

The only issue is if both options are out of the money at expiration I will have a naked futures position without a stop at the end of that day's session. I guess if the market hasn't moved in a decisive manner up to that point then maybe it won't suddenly, but I need to be up in time to put on a new options position.

The model is neutral for Monday. The different predictors point in different directions.

Another idea I had is that it is easy to adjust the ASX200 index for franking credits. S&P have a franking credit adjusted index but it only goes back to 2011 and has some weird features, like only reinvesting the dividends once a year. If you get the monthly values of the total return or accumulation index - which includes dividends but not franking credits and the price index which is without dividends, you can calculate the monthly dividend yield. The dividend yield can then be grossed up for franking - this will exaggerate franking a bit as some companies pay unfranked dividends. The return including franking credits matches the MSCI World Index gross total return since the financial crisis in 2009 very well:

My performance is also given pre-tax includes estimated franking credits. A major reason why I am lagging the index is presumably management fees.... You can see though that my returns about match the ASX in the last five years, despite the drag of management fees. This is by investing more in funds that do generate alpha. The black line is a simulation for the "target portfolio".