More Asymmetry

A couple of days ago I posted about the asymmetry of market returns capture by the target portfolio. The portfolio captured less than 100% of the upside in the markets but almost none of the downside. The chart below, inspired by a recent paper from AQR, shows the Bitcoin trading model's daily returns compared to the absolute percentage change in the price of Bitcoin futures for that day:

The rising diagonal line are all the days when the model was properly aligned with market direction. The descending diagonal line are all the days where it was incorrectly aligned with market direction. The remaining cloud of points is where the model changed direction. Some of those days were winners and some very bad losers when the model ended up incorrectly with the market in both directions that day. For example, it was stopped out of a long position and entered a short and then the market rose for the rest of the day...

The fitted quadratic curve shows that for low absolute price changes up or down in the price of Bitcoin, the model tends to lose money. This is because of "whipsaw". There is a strong asymmetry in the response for large moves and so the fitted curve shows that the model captures increasingly more of the return the larger the move.

The results do conform to AQR's argument that returns to trend-following have been poor recently because markets haven't been moving enough.

2008 Summary

I will be doing a report for December, but after such a financially disastrous year, I'm not in the mood for a detailed analysis of the numbers for 2008 as a whole. In US Dollar Terms we lost more than half our net worth and in Australian Dollar terms more than 40%. These results were partly due to the general decline in the markets and partly due to me not understanding the scope of the crisis and re-equitizing when only part of the decline was complete. I thought the collapse of Bear Stearns was the peak of the crisis. I was very wrong on that score. If we'd kept the conservative stance we had at the beginning of the year through to October or November we would be in a pretty good situation now with maybe a 20-25% decline in net worth in USD terms I think. Maybe better. In Australian Dollar terms we might have been down just 10% or so.

Some of the damage is permanent in realized capital losses and some is hopefully temporary due to currently depressed asset values. We're looking at realized capital losses of $A71,000 so far this year, with about $A14,000 of realized capital gains partly offsetting that. At least we won't be paying any capital gains tax any time soon :)

I started the year trying to be a short-term trader using my quantitative models for predicting short-term market direction. While I am convinced the models have some validity I found it very difficult to trade on their basis both due to being based here in Australia with most of the market action occurring overnight in US markets and my general problems of discipline in trading. I may still look to work with someone else in implementing the models to run a managed futures fund. Though given the Madoff Scandal there is likely to be less interest in blackbox models now. I'll return to look at these again once I have a couple more academic papers submitted. If you are a fund manager and are interested in working with me on this let me know.

Now at the end of the year I've moved much more towards an asset allocation/rebalancing approach to investing with limited market timing. I'd still expect to reduce exposure as the market rises and more so if the yield curve inverts. But I'd re-equitize much slower in any future market slump and never get as leveraged as I did this time around.

The year ended somewhat positively with what seems to be a gain for the month in USD terms though at the moment it looks like we lost in AUD terms. There were some positive signs also on my career front with an upcoming screening interview at a university and I'm getting my research back on track and now have two papers under review at academic journals and more in progress. Having an active research "pipeline" is important in getting an academic job at a good university. The two personal highlights of the year were getting married and visiting China and Hong Kong for the first time. My mother and brother visited us in Australia - my mother's first visit back here since she left more than 45 years ago and my brother's first visit to the country of which he is a citizen. Another positive personal thing is that in the last couple of months I have gotten back to doing a bit of cycling. Hopefully I can lose some weight in the coming year. We also bought Snork Maiden a bicycle and we've been on a few short rides together.

Asynchronous Diagonal Put Spread Trading

I'm now at the second stage in my new U.S. based options trading strategy. I'm trading QQQQ options at the moment. I'm short 4 June $48 puts and long 4 September $49 puts. This is a "diagonal put spread". A horizontal spread is the same strike in different months and a vertical spread different strikes for the same month. The asynchronous bit means that I didn't trade the two options at the same time, but allowed the market to move in between the two trades.

We know that ">put selling is extremely profitable. We also know that it can be extremely dangerous. For example, Victor Niederhoffer sold a large amount of out of the money S&P 500 puts in 1997. When the market fell the value of these puts rose partly due to the fall in the market and also due to the accompanying rise in implied volatility and his fund got a margin call that blew it up. This is despite the puts still being out of the money at that point. So I was interested in selling puts, but wary of the danger, and that led to the development of this strategy which I am beginning to implement. Another reason that I am interested in trading options, is that gaining from the time erosion of sold options means that I have to be less accurate in my trading in order to make money than is the case with futures. This is useful as I can't watch the US market all the time.

This is the strategy:

1. When my trading model gives a buy signal I short just out of the money puts for the current month. These puts have the largest time value of the current contract and the current month has the fastest time decay of all months. It makes no sense that I can see to ever sell anything but the current month (especially when spreads and commissions are low). I only sell a small number of puts - only as many as I am willing and able to buy stock in the event of being exercised at expiry. This is the first safety provision.

2a. Hopefully the market rises and the value of my puts falls rapidly. When my trading model generates a sell signal I buy a put. This put is for at least 3 months out - which has half the theta - sensitivity to time erosion. 6 months out has a third of the sensitivity but the bid-ask spread can be sufficient to negate the advantage. This means that the spread gains value over time ceteris paribus. The later option also has a greater vega - is more sensitive to volatility - which tends to increase when the market declines. Ideally, I buy this option a little in the money, but even if it is out of the money the delta (sensitivity to market movement) of this option is greater than that of the sold option. These differences in delta and volatility mean that if the market declines the spread increases in value. By not buying back the sold option I continue to gain from its time erosion and to save on commissions and spread - if the market is trending upwards it will likely expire worthless. It also partly hedges my bought put if I'm wrong about the market going down.

2b. If the market goes sideways, I just wait for the sold option to expire worthless.

2c. If the market goes against me - the model is wrong and I buy a later option to create a calendar spread to protect me against a crash and potentially benefit from rising volatility. When I think the market has bottomed, I'd probably just sell the bought option and redeem the sold option closer to expiry unless it was now very much in the money and had little chance of expiring worthless.

3. If 2a happens and we now have a diagonal spread, I wait for the model to generate its next buy signal. I then sell the bought put and hold the sold one and either it expires worthless or I buy it back closer to expiry. If the market has risen considerably since I first wrote the put, I will consider buying it back and writing a new one at a higher strike. And if the market has declined considerably, buying it back and writing a new one at a lower strike.

We are now just past 2a with the value of the spread increasing as the market declines.

I think this strategy cleverly exploits the advantages of put selling while mitigating some of the risk, exploits my model, and takes into account my location far from the U.S. time zone. I still need to be available either at the beginning or end of a US market session to make trades when neccessary.

Trading options in Australia has several disadvantages:

a. Very big contract sizes for the SPI futures options. I could write options more out of the money to reduce the risk, but that has Black Swan (i.e. tail event) risk.

b. The SPI futures options only exist for each quarter - my strategy is far more effective with monthly options.

c. There are also "XJO options" traded on the ASX 200 Index on the ASX. These are for 40% of the size of the SPI contract. Still big for me at the moment, but more manageable and these are monthlys. Problems with these are the very wide bid-ask spreads - SPI futures options have a tight spread (though you can't see bid ask quotes on IB's TWS...). If can get Interactive Brokers to approve me to trade them,* I can get low commissions - compared to CommSec's very high commissions. It's likely I will try trading these at some point.

Another possibility is futures options on the Nikkei traded on the Osaka futures exchange. These are also a big contract size but they are monthlies. The spread is 10 points, but so is the spread of the underlying contract. Commissions with IB are low and I already have the trading permission.

* I signed up for my account in the US and US residents are banned from trading options on foreign exchanges by the SEC. Futures options are OK because they are regulated by the CFTC. I've told IB I now live in Australia but regular Australian options are still blocked.

About as Long as I'll Get

The market started to rise tonight a day or two ahead of what my model was projecting. This recent correction was pretty shallow. So after re-running the models I bought every stock on my buy list in the US:

RICK, NNDS, SHLD, AAPL, GOOG, IFN, BWLD, HCBK, XLF, LUV and LCC puts

This is about as long as I'll get. Estimated portfolio beta is 1.2 and borrowing is 30% of net worth. Going forward I aim to gradually reduce the margin loans. RICK is not cooperating. NNDS is doing nicely since I bought back in.

P.S.

This post partly explains why. My model indicators also show a potential large rally ahead.

SPI Trading: Day 2

I planned to do live trading again today. My idea about what would happen at the open was correct - A pop up followed by a decline, but I missed the shorting opportunities, the first time because of caution and the second time because of inattention. Then I tried some simulated trades and lost three times in a row (-$10, -$185, -$225). No disasters, but I knew these were low probability trades, which is why I didn't try them with real money.

On the investing side, we are getting close to the potential buy point. The model is signalling Thursday as the beginning of an uptrend in NDX, SPX, and AORD. The Wednesday effect means that this could be brought forward. However, if we get into oversold territory then all bets are off until the market decides to reverse itself. The lowest my proprietary stochastic for the NDX can go on Tuesday is 23.4. But then even a decent 15 point bounce on Wednesday would still leave us oversold. But if the market falls less on Tuesday then even a small bounce will keep us out of oversold on Wednesday and then the reversal would be intact for Thursday.

NDX is getting close to the January lows but, based on the triangle formation since the January low, we should expect a low of 1610-ish. SPX is further from its lows. Stochastics on the weekly charts have also crossed and are now moving down, which is a necessary precondition for a meaningful low. The All Ordinaries is behaving very much like the SPX.

Bottom-line - if the market only sells off a little on Tuesday expect a bounce starting Wednesday. This week will not be the ultimate low. A strong sell-off on Tuesday is likely to result in the current downtrend producing the ultimate low for this bear market.

Short Term Trends: Asia vs. America

Though my US (NDX and SPX) models switched to short at Friday's close, the Australian and Japanese models are still long with no possibility of a short for several days in all likelihood. My guess is that this divergence means global markets will go sideways this week. In fact it is easy to see that if the indices remain unchanged for the next week, Australia will shoot up into the overbought zone as defined by the daily stochastics, while in the US the stochastics will decline. This isn't based though on any historically similar period - I probably should look into finding examples. I'd still expect plenty of volatility this week. But if this happens it will be another strong point in favor of the bottom being in.

The Australian market is closed today for Australia Day (which commemorates the arrival of the first British settlers in Sydney in 1788). But I'll be following Japan and the US futures to see what happens.

Game Plan

Next Wednesday's FOMC meeting is one day beyond the forecast horizon of my model. At this stage the model is saying to be long till the close Monday or maybe Tuesday. Being long the FOMC meeting looks to be a high risk proposition. I am thinking to stay long through Tuesday and around Tuesday's close sell a lot of my shorter term trades and hedge the rest of my portfolio with SKF. So I would go into the FOMC meeting with a market neutral portfolio that could outperform if the Fed doesn't cut rates. After this week's emergency 0.75% rate cut it's hard to imagine them cutting again, and yet market participants seem to believe that. I'll also go into IBKR's earnings release Thursday afterhours US time long 200 IBKR. I'll probably sell half my position then either way, good or bad. It worked last time.

At this point we've bounced around $US20k off the bottom.

First Real Trade of 2008

The slow stochastics crossed over on NDX so the model is long at the close. My models have been doing very well recently. I now have four: NDX, SPX, Nikkei, and All Ordinaries. After integrating the three separate econometric models into one it's now easier to maintain models of more indices. Anyway, I sold one BZH put (I keep one) and bought 400 QQQQ, 200 RICK, 200 SBUX, and 100 IBKR. The idea is to put on a more diversified model based swing trade. For some reason also stock feel safer than futures. I also went long NQ in my papertrading - the short trade I closed would have made $3000. I'll continue paper trading Australian and Japanese futures too. Barron's recommended Starbucks at the weekend and I am also playing a bounce in the restaurant stocks idea. I shouldn't have sold all of my RICK and planned to buy back at the next low. So here it is. Same with IBKR. I'm accounting all of these as trades, though, rather than investments as I don't think this is the market bottom yet. I'm wary of Starbucks as an investment too. The stock looks cheap based on earnings and their growth rate. But those earnings are only growing through massive capital investment. As a result, SBUX's free cash flow is a fraction of earnings. And it is free cash flow rather than earnings that we should use to value stocks.

P.S.

I got a lucky break on this trade - right after the market close SBUX announced they are bringing back Schultz as CEO and undergoing a restructuring. The stock is up 8% last time I looked.

Move is Complete

Snork Maiden's stuff arrived today, finally, after quarantine delays. They found some insect in the container. Now our apartment looks like a mixture between a garage sale and the tomb of Tutankhamun.

I've found a way of integrating the three separate stochastic trading models I was running. So in future I will only have to run one model for each index which will make things easier. I'm still looking for the optimal way to use the output so that decisions could be of a black-box type with no skill involved.

Neural Networks

Necessity is the mother of invention. In the last couple of days I've taught myself how to use neural networks to forecast time-series. Very helpfully, the econometrics package I am using - RATS - has a built in neural networks module, which is pretty easy to use. Reading some online papers by econometricans got me to see through the rather esoteric language used to understand that these are basically a specific type of non-linear regression model that isn't too far from some of things I've used in my academic research. I experimented with trying to forecast the stochastic oscillator but couldn't get anything better than I could with simple time series models. Now I am "training" a network to forecast changes in the NASDAQ 100 index using all the indicators that my existing model produces. The problem with the existing model is that there are several different indicators (there are three actual time series models for each index). Sometimes the direction is clear but a lot of the time, different indicators point in different directions. I've gradually developed ad-hoc rules for how to interpret the indicators in different situations but it is then very easy to second-guess and make mistakes. The neural network finds the best rules it can given the structure of neural net (explanatory variables, number of neurons etc.) and provides a simple long, short decision. So far the results are extremely good, but this is early days. Using this approach would add another layer of models to run every day but reduce the amount of time needed observing the market etc wondering what to do and whether the right decision was made.

In my simulated trading this week I did really well trading the Australian stock index futures (SPI) but not well trading the Nikkei. I think though I am beginning to get the hang of how to day trade the Nikkei.

Overall, it feels like I made progress this week, which is good.

Short the Nikkei

I've now estimated my model for the Nikkei and it is pretty successful in backtests. The model says to look for short-trades in the Nikkei on Monday. I probably won't start trading it though till Tuesday due to a couple of meetings I have during the day on Monday. Perhaps more on one of those meetings with someone interested in trading and modeling after we meet. A lot is happening this week actually. Should be interesting. I'll keep you informed.

P.S. The US models would suggest to switch to long Monday except that even if the indices were unchanged they would fall deeply into the oversold zone, which says to stay short. So only a very strong rally on Monday would actually show to get long. There is a reasonable likelihood of a positive gap up at the open Monday, that might be what you need to get long.

Volatility

I'm now turning to looking at volatility of stock prices and seeing if it could be a useful addition to my trading model. More sophisticated investors and traders and familiar with measures of stock price volatility derived from the implicit volatility expressed in options prices according to the Black-Scholes options pricing model. The VIX and VXN are the best known of these and measure the volatility implied by options on the S&P 500 and NASDAQ 100 indices. But one can also measure volatility directly from stock prices. The most straightforward would be a standard deviation of changes in the index. The problem with this and the reason why the options based indicators are popular is there will be a different result depending on how many observations are used to compute the standard deviation. This indicator also doesn't address intraday volatility. A measure of intraday volatility is Average True Range. True range measures the range of prices from previous close to close (and therefore includes any gaps in the range). ATR is simply an exponential moving average of this. A problem with this indicator is it is dependent on the level of prices. Of course we can simply divide true range by average price over the day to get a unit free measure of the daily range as a percent of price. The chart tracks a five day exponential moving average of this latter indicator.

The late 1990s and early 2000s were far more volatile than today with volatility peaking with an average of a 9% daily price range over the a five day period. As the post 2002 bull market took off volatility declined to very low levels and has now begun to re-emerge but not to the extent seen several years ago.

Why might volatility be interesting?

1. While mainstream finance theory claims that it is not possible to forecast stock prices (this is true though direction of stock prices may be forecastable) it is believed to be possible to forecast volatility in the short term. Typically ARCH (autoregressive conditional heteroskedasticity) time series econometric models are used. Being able to forecast volatility is a big advantage obviously in option trading and a reason I mostly avoid option trading except using deep in the money options as proxies for margined stock or futures. Volatility is not a component of futures prices which makes trading them a lot easier.

2. Stock prices are far more volatile when declining than rising. Market tops are more commonly characterised by narrow trading ranges that finally fail than by volatile "blow-offs". Market bottoms typically show violent intra- and inter-day fluctuations. If one forecast rising volatility - declining prices might also be associated with that forecast. Of course it makes sense that volatility is higher with declining prices -a rise in volatility implies a rise in risk and higher risk implies that lower prices are optimal - investors should pay less for a given amount of earnings with higher volatility assuming risk aversion.

At least volatility might explain some things about stock price behavior that my current completely price based model does not. So I'm going to do a bit of research on this. My first problem though is deciding on an appropriate indicator of volatility.

Effect of Exchange Rate Fluctuations on Returns

My recent net worth reports have shown huge fluctuations due to the volatility in the Australian Dollar-US Dollar exchange rate. Returns are strong in US Dollar terms when the Australian Dollar is rising - even though this is making us poorer in Australian Dollar terms. Each month I calculate the contribution to investment returns from the change in exchange rates under the heading "Forex" in my income and expenditure table and my table of returns on individual investments. In the last few days the Aussie has plummeted resulting in strongly negative investment returns for the month to date. This table shows just how much difference changes in the exchange rate make:



Stripping out the exchange rate results in lower average returns for the year so far (12.8% vs. 20.7%) but greatly lowered volatility and hence a higher Sharpe Ratio, which is a measure of the excess return (above a 5% hurdle in this case) divided by the standard deviation of returns. Both Sharpe Ratios exceed those for the MSCI and SPX total return indices. The SPX has risen less than the 5% hurdle so far this year (3.1%) and thus has a negative Sharpe Ratio. The MSCI has returned 12.8% at this point. A large part of that return is due to the fall in the US dollar. This is a global index measured in US Dollar terms. So really I'm doing neither as good, nor as bad as it might seem. I'm probably really beating the MSCI but not by as much as the crude numbers suggest. I've had two negative months - but both have come in the second half of the year, which has made me feel a bit despondent but the two indices have had four or five negative months.

Here is the same data for the more visually oriented:



BTW I haven't seen any comments in the personal finance blogosphere (obviously there's plenty on trading blogs) so far about this month's so far sharp fall in the indices. I guess it will come soon.

Following up from yesterday's blog. Actually, the model has been doing fine this month so far with only one stop-out so far (Friday). But I've been scared to get back on board due to its poor performance from the beginning of September to 2/3 the way through October. There were heaps of stop outs in late July and early August too. I wish there were a futures contract smaller even than the NQ (NASDAQ E-Mini) and I would be trading it overnight my time (US day time).

Model Facing a Crisis

My trading model has been performing poorly in recent months - the model is frequently stopped out - the market moves more than 1.25% in the opposite direction - far more than in other periods in recent years. As a result I have low confidence in placing trades according to the model. I've been in a desperate search to find a way to improve the model, but haven't found anything better than what I already have. The question is whether this is a temporary phenomenon or are the markets changing in some fundamental way? In the meantime, the best I can do is use the model as a guide to doing some intraday (or intranight) trades. I haven't done many of those recently either. I'm going to experiment a bit with some nonlinear model ideas, but I don't think they are likely to yield anything either.

Optimally Using Multiple Indicators

I've tried estimating a single model for both the NDX and SPX. The rationale was that taking itno account the correlation between the indices would use more information and, therefore, provider better signals for trading both indices. The results though were disappointing. Almost all indicators were almost exactly the same as those generated by my current individual NDX and SPX models. One exception - my furthest forward forecast for the SPX was severely degraded. Of course, this model fits the data better than my individual models. But it generates equal or poorer indicators. This is the usual situation in technical analysis. A simple moving average is not a good statistical model. But might be useful as a technical indicator.

But I am finding that using the raw predictions of the SPX and NDX models together is already leading to better decisions (BTW the model is long with Thursday being the likely start of the next decline). The next research exercise is working out how to use the two model signals together in an optimal way. For example, is it best to only take trades when both models indicate the same direction? It is easy to backtest this in an Excel spreadsheet (All my backtesting is done in Excel - I don't use anything more fancy - specific trading software wouldn't be able to run my model anyway).

P.S. I ran the tests - only taking a trade when both models (SPX and NDX) give the same signal actually makes NDX trading performance worse. Something more subtle is needed...

Trading Update and Travel Finance

Fifteen minutes is a tricky thing. It caused me to miss Apple's spectacular post earnings rise yesterday. But I still captured some of the effect by being long QLD (levered NASDAQ 100 ETF). I'm now beginning to regularly update both NDX and SPX models. This provides more information and makes it easier to make decisions when just looking at one index leaves one rather uncertain about what to do. Amazon earnings have pushed the market down after hours. The model is long though. After a large positive gap opening up on Tuesday a negative gap for Wednesday is likely. A good trade idea is playing the gap closing.

I'm now going to resort to a withdrawal from my Australian margin loan as the quickest way to get the money for the car and then juggling things around afterwards. I am supposing these various cards have limits on foreign transaction size which is what is causing them to be rejected. Snork Maiden's cards were now rejected by the bank for making a cash advance too. Beware if you are planning on travelling overseas and using credit cards for big bills. Or get some super-platinum card that you know won't have a problem. Just make sure it is widely accepted (i.e. not Discover or something, even American Express is less widely accepted than Visa and Mastercard). Maybe travellers cheques still have a role to play?

P.S. Australian margin loan money is on its way to us - we should have it Friday and the car soon after.

The Secret of Technical Analysis

I understand technical analysis to be any method that attempts to predict market moves based on past price and volume action rather than fundamentals. This includes the use of charts and also more sophisticated modelling. Most finance academics believe that securities follow simple random walk paths and technical analysis cannot predict anything. Now, much technical analysis probably isn't much use, its practitioners haven't tested the trading results based on it in a statistically valid way. The reason many traders probably make money is the use of stops. They stop their losing trades before they lose too much and let the winners run. Trend following approaches are similar. You will hear this advice very often when you start to study trading. In this case entry points can be more or less random. The profit-making assymetry is all in the stops.

In the last few days I've been researching various ideas I've had for improving my trading models. So far I haven't found anything better than I'm currently using. Some of the models fit the data better but aren't any better for trading. In fact they are worse. This is the secret. Models that fit the data well and have high levels of statistical validity are often not much use for trading. The type of models that no self respecting econometrician would choose are actually the best for trading purposes. This a major reason why academic finance rejects technical analysis in my opinion. The models they optimize to the data aren't actually useful for trading. But it is non-optimal models that can actually generate profits.

Allocating Cash

We are about to receive a refund from Snork Maiden's employer of moving expenses. They've agreed to pay more than originally proposed - a total of more than $A8,000. We will now have about 8% of net worth in cash outside of trading accounts. This is while we are borrowing almost 16% of net worth on an Australian margin loan at just over 9% interest. We only have about 3.5% though in Australian Dollars cash. The options are to:

1. Put most of this refund into a high interest money market account (about 5.5% interest) as an even larger cash buffer than we currently have (about $A7,500 in there currently and dividends and mutual fund distributions pay into this account).

2. Use it to reduce our margin loan. This has a higher certain return. We can always withdraw money from this account later though this probably requires sending a fax.

3. Transfer it to the US, buying US Dollars on the assumption that they are undervalued. This is possibly a high return (4.5% interest plus or minus change in value of the US Dollar) but risky.

Maybe we should do a little of each?

I still haven't placed any trades since we moved here. I now have the model up and running. In September and so far in October the model has underperformed the market. It doesn't do well in strongly overbought rallies as it has a somewhat bearish bias. So this has been as good a time as any not to trade. The last couple of days the model has been long and correct and the potential gain was $US400 per NQ contract. I didn't trade because the NDX seems exceptionally overextended relative to Bollinger Bands (i.e. the index is beyond 2 standard deviations from a moving average of the index) and my older "autoregressive model" is indicating a turning point is near. The model was short last Friday and would have been stopped out. The index rose 2.1%. I am waiting for the first good short opportunity. I have been doing some work on my modelling - continuing my earlier attempts to see if I can get a better edge in placing overnight trades (our daytime). I've come up with some regularities but nothing that seems reliable enough for systematic trading. The overnight sessions are less volatile but less correlated with the model than the intraday sessions. The two - overnight and intraday - have little correlation with each other. Perhaps it is best just to blindly place trust in my model and follow its signals. If I can actually do that.

Beginning to Relax

After the first week here we are beginning to be able to relax a little. We saw Parliament House, Floriade (and annual Spring flower festival in Canberra), the National Capital Exhibit (explains history of Canberra), and a stroll in Commonwealth Park along the shores of the lake with views to the Brindabella Mountains (picture of the Brindabellas above), and sat at an outdoor cafe in Garema Place. Canberra is cold in the shade and warm in the sun at this time of year.

I've also been testing a few trading modeling ideas. One result is that the correlation between the returns of the overnight session on the NDX and the intraday session is only 0.03. The overnight session has a correlation with the previous intraday session of -0.03. These are low correlations. Yet both have much higher correlations with my model's direction. Trading both sessions is a good diversification strategy. The overnight session (daytime in Oz) has half the volatility of the intraday session. A strategy that uses my model for market direction and trades twice a big a position overnight as intraday has a higher Sharpe Ratio (return/standard deviation essentially) than a strategy that always has the same size position. It has lower returns, but can be leveraged up more. I haven't yet computed Sharpe Ratios for the separate sessions individually. Probably something I should do. I also experimented with different stops. There is a risk-return tradeoff there too. A 1.25%-1.33% stop has maximum return for the NDX. A 1% stop strategy reduces the beta of the strategy to the index to zero (from about 0.15) but reduces returns (alpha is constant). Going to a 1.5% stop and above reduces alpha. In fact a no stops strategy in the last two years has a higher Sharpe Ratio than 1.5% or 2% stops. The key thing here is a setting a stop at a point where if the index reaches that point it tends to run away to the upside. The market often improves after going 1% in the wrong direction. It has much less of tendency to improve after going 1.25% in the wrong direction.

Wednesday Effect

The model was signalling to go long tomorrow. However, today is Wednesday which tends to be an up day in the markets. I was just searching for the link where I read about that statistically significant effect but I can't find it. Anyway in the last few years the market tends to go up on Wednesday. So I added a rule, that if the model is short on a Wednesday and Thursday is predicted to be long then I should go long on Wednesday. It improves the results. Just going long on Wednesday willy nilly, however, reduces returns. The reason for this is that on about half of all Wednesdays the model will be long anyway and that is a substantial part of the Wednesday effect. On the other half of Wednesday when the underlying model is short it is sometimes right and sometimes wrong. The times it tends to be wrong is when the signal is turning to the long side at the end of Wednesday anyway.

PS 4:25PM

The Wednesday effect obviously didn't work today though early in the day it looked like it was and I went long and made a profit. Then I went short, ditto. Then I went long near the close and the market kept falling. Then my internet connection went haywire :( I was stuck in the losing position and couldn't get out of it. When things were back to normal I was down more than 10 NQ points. There was a post-close bounce of sorts... deciding what to do now.