{% extends "backcast/base.html" %}
{% load humanize %}

{% block title %}Backcast: can any model beat the naive forecast?{% endblock %}
{% block masthead_search %}{% endblock %}

{% block content %}
<section class="intro">
  <h1>Can any model beat the naive forecast?</h1>
  <p class="lede">
    Pick a stock. Backcast goes back over its last two years of trading days, asks {{ rivals|apnumber }} models to
    forecast from each one using only the prices before it, and checks whether any of them did better than the
    simplest guess there is: the price later will be the price now.
  </p>

  <form class="query" action="{% url 'home' %}" method="get" role="search" novalidate>
    {% include "backcast/_combobox.html" with field_id="symbol" label="Ticker or company" value=symbol describedby="symbol-error" invalid=error required=True %}
    <fieldset class="segmented">
      <legend>Forecast</legend>
      <div class="segmented-options">
        <input type="radio" name="t" value="price" id="t-price" checked>
        <label for="t-price">Price</label>
        <input type="radio" name="t" value="volatility" id="t-volatility">
        <label for="t-volatility">Volatility</label>
      </div>
    </fieldset>
    <fieldset class="segmented">
      <legend>Forecast horizon</legend>
      <div class="segmented-options">
        {% for h in horizons %}
          <input type="radio" name="h" value="{{ h }}" id="h-{{ h }}"{% if h == horizon %} checked{% endif %}>
          <label for="h-{{ h }}">{{ h }} day{{ h|pluralize }}</label>
        {% endfor %}
      </div>
    </fieldset>
    <button class="button button-primary" type="submit">Run backtest</button>
  </form>
  <p class="field-error" id="symbol-error" role="alert">{{ error|default:'' }}</p>

  <p class="examples">
    Or start with
    {% for example in examples %}<a href="{% url 'backtest' example %}?h={{ horizon }}">{{ example }}</a>{% if not forloop.last %}, {% endif %}{% endfor %}.
  </p>
</section>

<section class="about" aria-labelledby="about-title">
  <h2 id="about-title">How a backtest works</h2>
  <p>
    On each of the last {{ origins }} trading days, every model sees the prices up to that day and forecasts the
    close 1, 5 or 20 trading days later. The forecast is scored against a price the model never saw, and nothing is
    tuned on the test period, so the error is the one you would have had at the time.
  </p>
  <p>
    A model can win on average and still just be lucky, and the more models you try, the likelier that gets.
    Hansen's SPA test allows for both. The <a href="{% url 'method' %}">method page</a> covers the rest.
  </p>
  <p>
    Across 498 stocks in the S&amp;P 500, no model beat naive by enough to matter. The
    <a href="{% url 'findings' %}">findings</a> have the numbers and the data.
  </p>
</section>
{% endblock %}