Many have understood the downsides of historical dervied indicators, that is, having no predictive value. People started to do forecasting with many sophisticated models – those that believe volatility is persistant, cyclical, mean reverting, and conditional variance on sound economic rationale underlying the move (MDE), or simply regressing on itself with some memory of its past and a long term average volatility(GARCH). Such model fitting has huge model risk if model attempts to “chase history” rather than predict, with higher orders of moments and returns such that a simple GARCH(1,1) or EWMA forecast is more reliable if one were to compare the in-sample forecast and out-sample forecast error.
But, but, the most important point here is… out-sample don’t use any value from today. It only uses historical value. There will be no hindsight biases, no overfitting. Whereas in- sample error incorporates information from today and allows it to chase and overfit. MDE with good underlying economic factors (e.g. inflation, yield curve, credit risk, risk premium, earnings growth revision are sound for equities; forwards, rates differentials, options risk reversals, implied volatity and its term structure are sound for currencies, swap, credit spread are sound for fixed income) perform better than EWMA, but EWMA is used more often for those with no access to such financial datasets. That said, we do see many traders in a small team with no quants having moving averages on their charts, and this explains why prices have ‘reflex behaviours’ at moving averages bouncing off and coming back, flirting with these key levels, till market is tire of its action before it starts its perpetuated ascend or descend.
But little did people know that, before moving average concepts were used, people have ideas of what “unit of measurement” defined a big or small move. They are also clearly aware of the downsides of moving averages, that exhibits ‘ghosting effect’ and sudden fall in averages where N+1 days past and it experienced an unprecedented huge move N+2 days ago that causes market turmoil (i.e. >4*STDEV where by the chance of it happening, its considered a ‘black swan’). Although it defies the concept of efficient maket hypothesis, where information are known by most and act upon immediately as data is released and digested, some had ideas of conditional mean. To kept it simple, they simply condition it that people memory were short and only can remember a short period of time. That led to pivot levels as one key prices for traders so they have more nearby and extended levels to carry out their job for their clients. Being a quantitative analyst, I did a discrete (treating price range within pivots as category) markov chain as well as simple binning to see how prices react near these levels.
Source: Bloomberg, R
Ideas of Elliot wave and harmonic patterns came out in the community from presistant observation of price actions. But the foundation was fibbonacci, built on wave strcutures, or some would call them ‘fractuals’. Most would simply retrace previous larger wave. But if one were to think that smaller waves now might evolved as blocks for larger waves later, and if one were to lookback, small waves will correspond to the larger wave. That meant that it not only a tool for getting value from historical key levels, but projecting key levels in the future. If one believe in fibbonacci, one will also believe in the idea that most things in life after many repetitions will converge to some place at some time before it extends to infinity. Refer to Mandelbrot “The (mis)behaviour of markets” for ideas of how I decompose wave structures into larger fractuals, and within each fractuals, small fractuals. Hint: I did a recursion with some Pk-Tr-Pk-Tr, Tr-Pk-Tr-Pk merging and ordering (smallest fractual is either of these zig zag) .
Source: Bloomberg, R
Source: Bloomberg, R
Besides extending pivots (conditional on short memory) average with range of most recent price moves, and projections and expansions of wave fractuals, more common to most traders are trendlines. More importantly, they use lognormal scale when prices exhibit volatility dependence on its price levels. It is another tool used with predictive value of where prices will have a dump-pump or flirting behaviour – only if underlying fundamental cycle is not strong to convince everyone in the opposite positions now to flip or close their positions, which led to some long holders seeing an opportunity to add on positions. Rule is that, have no assumptions of where the flirting will go, until the couple breaks up with a clear margin.
Source: Bloomberg, R
As of now, using Quandl price from Wiki Exchange Rates, my trendline algorithm have picked up a signal from GBPUSD.
Note: UK retail sales fell by the most to -1.9% MoM, while Teresa May is in the midst of securing the deal that she thinks would be the best for UK future and Article 50 is yet to be enacted. And don’t forget about the counter currency USD, Trump just took office and is set to unleash a string of protectionist policies such as border taxes. USD should strengthen if 3 rate hikes promised were materialised as the market believed, as seen from the yields breakout and almost doubled.
So why is key levels so important?
I say, fundamentals second, price first. There are many downsides to improper entries without at least some consensus of what is a ‘good’ price to enter. Entering at any random price ‘when you suddenly grasp the rationale of the trade idea’ is the first step to killing your account. One, your SL and TP are not aligned with anyone in the market, and you are stuck in the midst of fluctuations, where any upside or downside move is random and does not signal to you that you are right or wrong. There is a saying that, key prices will always be tested before anyone knows whether its worth defending for them, and levels are where positions are reversed most of the time. Entering at clear, unanimous levels signals to you that when the prices are a wide margin away from your positions, the market have gave up defending that level, and all will be waiting for price to head to the next nearest level. This gives a very sound approach to setting SL. After all, any levels is better than no levels. On a side note, while levels are important, counts of waves and what part of cycle you are entering is also equally important, as a strong cycle can violate important levels without looking back – usually, it will come back again for a final test. So do pay attention to extent of wave correction (or are they trend?)