1
$\begingroup$

I can understand for speech signals, words are correlated and therefore one should have a reason to believe that recurring NNs or LSTMs could predict by running some complex algorithm with weights and an activation function.

But for random digital signals like stock prices, why can the prices be predicted in any way? Tomorrow's price has nothing to do with those in the past.

$\endgroup$
1
$\begingroup$

According to the Efficient-market-hypothesis, you are right: the price movement of today does not affect the price movements of yesterday. However, this hypothesis describes an abstract assumption, which can be very useful for macroeconomic modeling but should not be considered as a comprehensive description of reality (The same applies to other economic assumptions, like that of the Homo economicus for example.).

In fact, there is a lot of criticism regarding the efficient market hypothesis. The Cryptocurrency-Bubbles and the Dot-com bubble are most probably the best examples that price movements in financial markets are not pure random walks in many cases, but interfere with behavioral psychological effects.

From my experience, I can assure you that many indicators are quite consistent: if you're trading in a growing market, the probability of an increase in prices is slightly higher than that of a decline (even though there will be a tipping point sooner or later). In addition, there are other indicators, such as volatility, which are relatively stable. Of course, you could try to manually determine all these indicators and include them as features in an ordinary DNN, but in practice, using an LSTM usually turns out to be easier and more accurate.

| improve this answer | |
$\endgroup$
  • $\begingroup$ good. can u give an explanation how LSTM predicts future values based on the history? may i understand it as having a short term memory thus relying more on more instant data? $\endgroup$ – feynman Mar 8 '19 at 3:06
  • 1
    $\begingroup$ Imagine that you predict prices in intervals of 15 minutes and that you found out that a strong increase in prices in one interval usually leads to a corresponding price correction within the next 45 minutes. If you don't use an RNN, you will have a hard time modeling this knowledge. You would have to include the dummy variable strong_price_increase_within_last_45_mins or something like that in your model. If you use an RNN, however, you don't have to include any new features to model that knowledge. Eventually, the RNN will find that pattern out by itself. $\endgroup$ – georg-un Mar 8 '19 at 11:42
  • $\begingroup$ why is there a corresponding price correction within the next 45 minutes following the strong increase? $\endgroup$ – feynman Mar 8 '19 at 14:02
  • 1
    $\begingroup$ This was an imaginary example to illustrate why the use of an RNN for price forecasting can be beneficial. I think I misunderstood your first comment. In a feed-forward NN, you don't have any memorization. But in an LSTM you do. You can imagine it as a NN in which every neuron remembers its state over multiple time intervals and forgets old states only with a certain rate. This means, instead of only relying on current data it can take data from the previous time intervals into account to find underlying patterns. This makes especially sense in time series data like price movements. $\endgroup$ – georg-un Mar 8 '19 at 15:11
  • $\begingroup$ but what if the corresponding price correction within the next 45 minutes didnt actually happen, what will the LSTM predict after the 15 min interval $\endgroup$ – feynman Mar 8 '19 at 15:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.